Compare commits

...

56 Commits

Author SHA1 Message Date
Nikhil Sonti
c02cdab095 fix: address balpha CLI dogfooding feedback 2026-04-27 15:39:39 -07:00
Nikhil
27834b1d31 fix: udpate readme (#829) 2026-04-27 15:27:16 -07:00
Nikhil
aa30eb3aaa feat: add balpha dogfooding CLI (#828)
* feat(alpha): scaffold balpha cli

* fix(alpha): address scaffold review

* feat(alpha): add balpha config

* feat(alpha): parse browseros profiles

* feat(alpha): import browseros profile

* feat(alpha): add browser launch helpers

* feat(alpha): add repo build and env pipeline

* feat(alpha): add process supervision

* feat(alpha): add balpha commands

* docs(alpha): document balpha setup

* fix(alpha): reuse dev setup script

* fix(alpha): address review feedback

* fix(alpha): normalize imported browser profile

* fix(alpha): use generic profile fixture names
2026-04-27 15:03:37 -07:00
shivammittal274
e045e34b73 fix(eval): switch weekly eval configs from Fireworks to OpenRouter (#827)
The 2026-04-23 weekly run had 42% of AGISDK and 46% of Infinity tasks
fail with `AI_RetryError: ... the service is overloaded` from Fireworks
(20 concurrent kimi-k2p5 streams across both runs at 10 workers each).

Switching to OpenRouter (which fronts the same Moonshot K2.5 weights
and falls back across providers) for the three weekly configs:
- browseros-agent-weekly.json
- agisdk-real-smoke.json
- infinity-hard-50.json

Model accounts/fireworks/models/kimi-k2p5 -> moonshotai/kimi-k2.5
(same weights, same 262K context). API key env var, base URL updated.

OPENROUTER_API_KEY is already wired into .github/workflows/eval-weekly.yml
and present in repo secrets — no GH config changes needed.

Orchestrator-executor configs and test_webvoyager left on Fireworks
intentionally; can switch later if needed.
2026-04-27 21:52:26 +05:30
shivammittal274
01d649da9a feat(eval): bring deterministic graders to dev + drop omnizon (#824)
* feat: deterministic eval graders (AGI SDK + WebArena-Infinity) (#664)

* feat: add deterministic eval graders (AGI SDK + WebArena-Infinity)

Two new benchmark integrations with programmatic grading — no LLM judge.

AGI SDK / REAL Bench (52 tasks):
- 11 React/Next.js clones of consumer apps (DoorDash, Amazon, Gmail, etc.)
- Grader navigates browser to /finish, extracts state diff from <pre> tag
- Python verifier checks exact values via jmespath queries

WebArena-Infinity (50 hard tasks):
- 13 LLM-generated SaaS clones (Gmail, GitLab, Linear, Figma, etc.)
- InfinityAppManager starts fresh app server per task per worker
- Python verifier calls /api/state and asserts on JSON state

Infrastructure:
- GraderInput extended with mcpUrl + infinityAppUrl for parallel workers
- Each worker gets isolated ports (no cross-worker state contamination)
- CI workflow: pip install agisdk, clone webarena-infinity repo

* chore: switch eval configs back to kimi-k2p5

* fix: register deterministic graders in pass rate calculation

Add agisdk_state_diff and infinity_state to PASS_FAIL_GRADER_ORDER
in both runner types and weekly report script, so scores show correctly
in the dashboard.

* chore: temp switch to opus 4.6 for eval run

* chore: restore kimi-k2p5 as default eval config

* ci: add timeout and continue-on-error for trend report step

* fix(eval): drop omnizon from AGISDK dataset (DMCA takedown)

evals-omnizon.vercel.app returns HTTP 451 ("This content has been
blocked for legal reasons / DMCA_TAKEDOWN"). All 5 omnizon-* tasks
fail grading with "Failed to fetch /finish endpoint: JSON Parse error".

Adds an EXCLUDED_WEBSITES set to the dataset builder and regenerates
agisdk-real.jsonl (52 → 47 tasks).

* fix(eval): correct Infinity port-assignment bugs

Two related bugs in the Infinity eval runner that cause silent port
collisions / fallbacks under parallel execution:

1. build-infinity-dataset.py emitted "app_port" but task-executor and
   the committed JSONL both read "app_base_port". Re-running the build
   script would silently make every task fall back to the 8000 default,
   ignoring per-app port assignments. Renamed the key to match.

2. task-executor derived workerIndex as `base_server_port - 9110`, but
   parallel-executor doesn't override base_server_port per worker —
   only server_url. Every worker computed workerIndex = 0, causing all
   parallel workers to spawn Infinity app servers on the same port.
   Threading workerIndex explicitly through TaskExecutor instead.

Also drops an unused app_name parameter from load_tasks().
2026-04-27 21:35:43 +05:30
Dani Akash
ddbb2cf492 feat(agent): composer attachments + server-side outbound message queue (#826)
* feat(agent): attach images and text files to chat messages

Adds end-to-end support for image and text file attachments in the chat
composer, with the staged files round-tripping through the OpenClaw
gateway as OpenAI-compatible content blocks and persisting in the JSONL
so they show up in the historical view.

Server
- HTTP client: new OpenClawChatContentPart union and a buildUserContent
  helper that emits multimodal content arrays when messageParts is
  supplied, falls back to the legacy string content otherwise.
- Service: chatStream takes an optional messageParts array and forwards
  it; BrowserOSChatHistoryItem gains an attachments field.
- JSONL reader: PiContentBlock learns the OpenAI image_url and Anthropic
  image source/data shapes; user messages now emit user.attachment
  events that the history mapper accumulates onto the next user item.
- Route: validates an inbound attachments[] (kind/mime/size/count),
  inlines text-shaped files as <attachment> blocks in the message body,
  attaches images via image_url parts. Replaces the immediate 409 on
  active monitoring session with a 30s waitForSessionFree(agentId) wait
  (registry now exposes onSessionEnd) so cron/hook contention does not
  reject a user-chat send outright. Returns 503 if the wait times out.

Client
- New lib/attachments.ts: validateAttachment / compressImageIfNeeded
  (canvas downscale to 2048px long edge, JPEG 0.85 re-encode for >1.5
  MB inputs) / stageAttachment / stageAttachments that produces the
  staged-attachment shape the composer renders and the payload the
  server accepts.
- ConversationInput: drag-and-drop, paperclip button, clipboard paste,
  staged attachment chip strip with thumbnails for images and a
  paperclip+name chip for text files. Send button enables on either
  text or attachments. Drop-zone overlay during drag.
- chatWithAgent forwards attachments[]; useAgentConversation.send
  accepts a SendInput shape and renders user attachments on the
  optimistic streaming turn via MessageAttachments / MessageAttachment.
- ClawChatMessage groups historical attachment parts into a single
  MessageAttachments strip, ordered before reasoning/tools/text.
- claw-chat-types adds an attachment ClawChatMessagePart variant; the
  history mapper emits attachment parts first and skips the text part
  when the user only sent media.
- AgentCommandHome forwards the new SendInput shape — home composer
  drops attachments at the boundary in v1 (the conversation page is
  where staging is most useful; carrying bytes through the URL bar
  is not sensible).

Limits: 10 attachments per message, 5 MB per image (post compression),
1 MB per text file, mime types png/jpeg/webp/gif and text/* +
application/json. PDFs and other binaries are deferred to v2.

* feat(agent): outbound message queue for chats while agent is mid-turn

Lets users keep typing and submitting messages while the agent is still
streaming a previous turn. Each press is appended to a single-flight
queue and dispatched as soon as `streaming` flips false; the queued
state renders as a strip above the composer so the user sees what's
pending vs. what's already sending.

- New `useOutboundQueue` hook owns the queue, the worker effect, and
  cancel/retry actions. Single-flight by design — a re-entrancy ref
  guard prevents two simultaneous dispatches when `streaming` flickers.
- Composer (`ConversationInput`) accepts optional `outboundQueue`,
  `onCancelQueued`, `onRetryQueued` props. When the queue is provided
  the send-button gate stops blocking on `streaming`; the spinner stays
  as the visual cue that the agent is still busy. Legacy direct-send
  callers keep the old streaming-blocks-send semantic.
- Renders an OutboundQueueStrip above the staged-attachment strip with
  per-item status (queued / sending / failed), a cancel button on
  queued items, and retry + discard on failed items.
- AgentCommandConversation wires `onSend` to `queue.enqueue` and routes
  the home composer's `?q=` initial-message handoff through the queue
  too, so it inherits the same single-flight serialization.

The server-side `waitForSessionFree` (added with attachments) and this
client-side queue together cover both contention sources: cron / hook
turns and back-to-back user sends. Persistence across reloads is
intentionally out of scope for v1 — losing the queue on extension
reload is documented as a known limitation.

* feat(server): server-side outbound message queue

Replaces the client-only React-state queue from 123ef21d with a
proper server-owned queue. Closing the tab is now safe — the server
holds queued messages and dispatches them through the existing
chatStream path the moment the agent's ClawSession status flips to
idle.

Server
- New OutboundQueueService (apps/server/src/api/services/queue) — per
  agent FIFO, in-memory. Subscribes to ClawSession.onStateChange
  through OpenClawService.onAgentStatusChange, and dispatches via
  OpenClawService.chatStream so attachments / history / monitoring
  all behave identically to the existing /chat route. The worker
  drains the SSE response server-side so the gateway run finalizes
  cleanly even with no client connected.
- Four new routes under /claw/agents/:id/queue:
  POST   /queue            enqueue
  DELETE /queue/:itemId    cancel a queued item
  POST   /queue/:itemId/retry  re-queue a failed item
  GET    /queue/stream     SSE feed of the per-agent queue state.
  Validation reuses validateChatAttachments and
  buildMessagePartsFromAttachments from the existing chat route.
- Singleton wired in apps/server/src/main.ts; shutdown on SIGTERM.
- New OpenClawService.getAgentState getter for the queue worker's
  pre-dispatch sanity check.

Client
- useOutboundQueue rewritten as an SSE-backed projection over server
  state. Public API unchanged so the composer still works.
- enqueue POSTs to /queue and shows an optimistic local entry until
  the server's SSE snapshot reflects it; local-only entries get a
  `local-` id prefix so cancel can short-circuit them without
  hitting the server.
- AgentCommandConversation watches the queue for sending items
  dropping out and refetches history so the new assistant turn shows
  up in the conversation view (the server worker streams the
  dispatched turn into OpenClaw without exposing per-turn SSE to
  the client).

Out of scope (documented in the plan as v2 follow-ups): disk
persistence (server restart loses queue), per-turn live streaming
of queued sends in the conversation view, and switching the
underlying dispatch from /v1/chat/completions to the chat.send RPC
(which would also fix the multimodal attachment routing problem).

* fix(server): outbound queue must reuse existing session, not spawn UUIDs

The queue worker was generating a fresh randomUUID() as the sessionKey
when the queued item didn't carry one — and the client wasn't sending
one. Result: every queued message kicked off a brand-new OpenClaw
session, orphaning the user's active conversation behind the new
"most recent" entry in sessions.json. The history endpoint then
resolved to the orphan and the chat appeared to disappear.

Fix is layered:
- Client (useOutboundQueue): forward the current resolvedSessionKey
  in the POST /queue body so every queued message targets the same
  conversation the user is viewing. AgentCommandConversation passes
  resolvedSessionKey into the hook.
- Server (OutboundQueueService): the worker now resolves to the
  agent's existing user-chat session when no sessionKey is provided
  on the queued item, via OpenClawService.resolveAgentSession. UUID
  fallback is now reserved for the first-ever message on a brand
  new agent — same semantic the existing /chat route has implicitly
  through the catalog of historical sessions.

No JSONL data was lost by the original bug (the prior conversations
are intact on disk); the orphan sessions just shadowed the original
in sessions.json.

* fix(agent,server): address PR review feedback for chat queue

- Tighten image data URL cap to base64-aware ~6.7 MB (was ~7.5 MB
  through `MAX_IMAGE_BYTES * 2`).
- Forward chat history from useOutboundQueue.enqueue so queued sends
  preserve conversation context like direct sends do.
- Match local attachment previews to server snapshots by id (not by
  message text), and prune the preview map as items drain.
- Pass an AbortSignal into chatStream so a queue shutdown cancels the
  initial OpenClaw handshake, not just the SSE drain loop.
- Track previously gitignored apps/agent/lib/attachments.ts (was caught
  by global lib/ ignore) so CI typecheck can resolve @/lib/attachments.
- Update server-api openclaw route tests to the new chatStream signature
  and the waitForSessionFree-based busy-agent path.

* fix(agent): dedupe optimistic queue entries for text-only sends

The localId↔serverId map was only populated when the message had
attachments, so plain-text sends left the optimistic local entry in
place after the server snapshot arrived — the user saw the same
message rendered twice in the queue strip.

* fix(agent): prune optimistic queue entry on POST ack, not just SSE

The server broadcasts the new queue snapshot before its POST response
returns, so the SSE handler often runs first — at that point the
localId↔serverId map has no entry for the new server id yet, so the
SSE-based dedupe path can't drop the optimistic local entry. Pruning
on POST success closes the race deterministically.

* fix(agent): hand off optimistic queue entry without a render gap

Pruning the local entry on POST success only worked when the SSE
snapshot had already overwritten it; if the POST response landed
first, the optimistic row disappeared for a frame before the SSE
snapshot brought back the server-keyed row, producing a visible
flicker. Gate the POST-side prune on the SSE snapshot already
carrying the server id, and rely on the SSE-based dedupe (now
guaranteed to find the localId↔serverId link in the map) to clean
up when SSE arrives later.

* fix(agent,server): client-generated queue id eliminates render flicker

The server used to assign its own UUID when an item was enqueued, so
the optimistic client row carried a `local-` id while the SSE snapshot
carried a server UUID — the client had to wait for the POST response
to learn the mapping before it could dedupe, and during that window
both rows rendered.

Now the browser generates the id, sends it in the POST body, and the
server uses it verbatim (falling back to a fresh UUID only if the id
collides with an existing item). The client collapses to a single
id-keyed list, so the optimistic row and the SSE row reconcile on the
same key from the very first render.
2026-04-27 21:31:03 +05:30
Dani Akash
711934555d feat(agent): enrich chat UI with tool activity, reasoning duration, and cost (#825)
* feat: pass per-turn cost and token data through chat history items

- Add costUsd, tokensIn, tokensOut to BrowserOSChatHistoryItem (server)
- Pass through from JSONL agent.message events in jsonlEventsToHistoryItems()
- Add same fields to client-side BrowserOSChatHistoryItem and ClawChatMessage
- Map cost/token data in mapHistoryItemToClawMessage()

Data flows: JSONL message.usage → server history item → API response →
client ClawChatMessage. Available for rendering in ClawChatMessage
component (message toolbar, cost badges).

* feat: add message toolbar with copy button and per-turn cost display

Add MessageToolbar to historical assistant messages in ClawChatMessage:
- Copy button copies message text to clipboard via MessageAction
- Per-turn token count (22.7K → 238) and cost ($0.003) shown as muted
  tabular-nums text on the right side of the toolbar
- Toolbar appears on hover (opacity transition via group-hover)
- Only shown when the message has text content
- Cost/token display only shown when data is available from JSONL

* fix: toolbar only on assistant messages, always visible, cost only

- Only render toolbar on assistant messages (not user messages)
- Remove hover-only opacity — toolbar is always visible
- Remove token counts (22.7K → 238 is meaningless to users)
- Show only cost as a budget signal ($0.003)

* feat: group all tool activity into single Task collapsible per turn

Replace flat tool rows with a single ai-elements Task collapsible per
assistant turn that lists every tool/MCP call in sequence.

Live streaming (ConversationMessage):
- Aggregate all tool-batch parts into one Task
- Title: "Working… (N actions)" while running, "Agent activity (N actions)" when done
- Default open while turn is in progress
- Wrench icon in trigger

Historical (ClawChatMessage):
- Group all tool-call parts into one Task
- Title includes failed count if any tools errored
- Default collapsed — expandable on click
- Tool name + status icon + error text per row

Both views show one clean collapsible per turn instead of N individual
tool cards. Collapsed reads "5 actions"; expanded shows the timeline.

* feat: include tool calls in chat history responses

Server: jsonlEventsToHistoryItems() now walks ALL events (not just
messages) and pairs agent.tool_use with agent.tool_result by toolCallId.
The resulting tool call list is attached to the next assistant text
message as toolCalls[]. Each entry includes status, input arguments,
output text, error string, and duration computed from event timestamps.

Client:
- BrowserOSChatHistoryItem gets optional toolCalls field
- Tool-call message part type gets durationMs field
- mapHistoryItemToClawMessage() emits tool-call parts BEFORE the text
  part (the order the agent produced them)
- ClawChatMessage Task view now shows tool duration in seconds

Result: historical messages now display the full tool activity
timeline grouped into the single Task collapsible per turn (designed
in step 3), instead of showing only the final text response.

* feat: render activity rows as human verbs sourced from tool registry

Tool calls in the chat activity view now read as sentences:
"Opened tab · news.ycombinator.com" instead of "browseros__new_page".

Server (tool-label-registry.ts):
- Curated verb override map for ~70 BrowserOS first-party tools
- Per-tool subject extractors that pull the meaningful argument from
  input (URL → host, query → quoted, element → ID, etc.)
- Generic fallback humanizes snake_case for any unmapped tool
- Strips MCP namespace prefixes (browseros__, mcp_)

Server (openclaw-service.ts):
- jsonlEventsToHistoryItems calls buildToolLabel for each tool_use,
  attaches label and subject to the BrowserOSChatHistoryToolCall

Client:
- Mirrored label module at lib/tool-labels.ts
- useAgentConversation tool-start handler computes label/subject
  from the SSE tool args
- ClawChatMessage and ConversationMessage render label · subject
  with foreground/muted styling, no font-mono
- ToolEntry, BrowserOSChatHistoryToolCall, and tool-call message
  part types all carry label and optional subject

* fix: drop meaningless tab N subject from page-read tool rows

Page IDs are internal numbers, not URLs. 'Took screenshot · tab 4'
tells the user nothing. Removed subject extractors for take_snapshot,
take_enhanced_snapshot, get_page_content, get_page_links, get_dom,
and take_screenshot. The verb alone is the right signal.

* fix: gate initial loading on historyQuery.isFetched not isLoading

The session and history queries are sequential: the history query is
disabled until session resolves. After session resolves, there's a render
frame where historyQuery.isLoading is still false (the query hasn't
been kicked off yet). isInitialLoading flipped to false during that
window, exposing an empty chat shell with just Task collapsibles and
copy buttons before the messages filled in.

Switching the guard to isFetched closes that window — the loading state
stays true until the first history fetch actually completes.

* fix: render historical messages immediately instead of through Streamdown's idle-callback debounce

Streamdown defaults to mode="streaming" which uses requestIdleCallback (300ms
debounce, 500ms idle timeout) and lazy/Suspense to optimize for token-by-token
live streams. For finalized historical messages this caused tool collapsibles
and copy buttons to paint while text bodies stayed blank for ~300-500ms after
load. Pass mode="static" + parseIncompleteMarkdown=false on the historical
MessageResponse so completed text paints in the same frame as the surrounding
chrome. Live streaming turns still use the default streaming mode.

Also collapse the redundant /agents/:id/session round-trip into the existing
/history endpoint (server already resolves the most recent user-chat session
when sessionKey is omitted) and tighten the initial-loading gate to stay true
across the render frame where the query is enabled but hasn't started fetching.

* feat: surface thinking duration on historical reasoning collapsibles

Server accumulates agent.thinking events per turn from JSONL and attaches a
single reasoning block (joined text + durationMs from first thinking event
to the closing agent.message) on each assistant history item. Reasoning
buffer resets on user.message alongside the tool-call buffer.

Client mirrors the type, emits the reasoning part before tool calls in
mapHistoryItemToClawMessage (chronological: think → act → answer), and
passes duration in seconds to <Reasoning> so the trigger reads "Thought
for N seconds" instead of just "Thinking" on collapsed historical turns.

* fix: read thinking blocks from the correct JSONL field name

OpenClaw stores reasoning blocks as {type:'thinking', thinking:'...'} but
the JSONL parser was reading block.text, so every thinking event was
silently dropped before it ever reached jsonlEventsToHistoryItems. As a
result the reasoning field on history items was always empty even though
the new accumulator was wired up correctly.

Also guard the client mapping: when durationMs is 0 (think + answer
emitted in the same JSONL line, no real elapsed wall-clock) pass
undefined to <Reasoning> so it renders the static "Thinking" trigger
instead of the streaming shimmer / "Thought for 0 seconds".

* fix: reset reasoning buffer on discarded turns and drop dead session hook

Two cleanups from PR review:

1. jsonlEventsToHistoryItems: when an agent.message is discarded (the
   "[Chat messages since your last reply" wrapper without a current-message
   marker) the tool buffers were already reset but the reasoning buffer
   was not. Accumulated thinking from the discarded turn would bleed onto
   the next assistant message. Reset pendingReasoningTexts and
   pendingReasoningFirstAt alongside the tool buffers.

2. useClawAgentSession, the AgentSessionResponse type, and the unused
   session entry in CLAW_CHAT_QUERY_KEYS became dead code after the
   session round-trip was folded into the history endpoint. Removed.
2026-04-27 18:29:15 +05:30
Nikhil
5125dffbf3 fix: sign limactl with VZ entitlement (#822) 2026-04-26 13:30:09 -07:00
Dani Akash
0035893f33 feat: dashboard API, JSONL reader, and OpenClaw observer for enriched home page (#810)
* feat: draft agent chat ui exploration

* feat: refine agent chat ui draft

* feat: remove outer frame from agent chat workspace

* fix: offset agent chat for app sidebar

* fix: simplify agent conversation shell

* fix: remove redundant chat header actions

* fix: unify agent conversation headers

* fix: tighten agent chat spacing

* fix: bound agent chat composer height

* fix: remove agent chat page inset

* fix: align agent header height with sidepanel

* fix: center agent composer resting state

* fix: anchor multiline composer controls

* fix: remove focus grid from agent home

* fix: remove redundant agent home header

* fix: constrain home agent composer

* fix: match home composer default posture

* feat: add openclaw chat history APIs

* feat: add claw chat history hydration

* fix: stabilize claw chat viewport layout

* fix: use conversation scroll base for claw chat

* refactor: split claw chat controller responsibilities

* fix: keep active agent turns in memory

* fix: normalize openclaw chat sessions

* refactor: use HTTP client for agent history instead of CLI client

Replace the CLI-based getChatHistory() call in getAgentHistoryPage()
with the HTTP client's getSessionHistory() from PR #795. This uses
the direct HTTP transport to OpenClaw's /sessions/<key>/history
endpoint instead of shelling out through the CLI.

- Add filterHttpSessionHistoryMessages() for flat-string content format
- Add normalizeHttpHistoryMessages() for OpenClawSessionHistoryMessage shape
- Update getAgentHistoryPage() to call getSessionHistory() via httpClient
- Remove unused getChatHistory(), filterOpenClawSystemMessages(),
  normalizeChatHistoryMessages(), and getTextContent()
- Update test mocks from cliClient.getChatHistory to httpClient.getSessionHistory
- Update MutableOpenClawService type: chatClient -> httpClient

* fix: fetch all session messages by iterating OpenClaw pagination

OpenClaw's HTTP history endpoint returns a limited page by default.
When called without a limit, only the first ~27 messages were returned,
causing all newer conversation messages to be silently dropped.

Add fetchAllSessionMessages() that iterates through OpenClaw's cursor-
based pagination (200 messages per page) until hasMore is false, then
feeds the complete message list into the existing BrowserOS normalization
and in-memory pagination layer.

* refactor: migrate chat history from HTTP gateway to direct JSONL file reads

Replace the HTTP-based chat history pipeline (BrowserOS server → OpenClaw
gateway /sessions/:key/history pagination loop) with direct JSONL file reads
from the host filesystem via Lima's virtiofs mount.

- Add OpenClawJsonlReader that reads session JSONL files directly from
  ~/.browseros/vm/openclaw/.openclaw/agents/<id>/sessions/
- Replace fetchAllSessionMessages() HTTP pagination with single file read
- Replace CLI-based listSessions() with sessions.json file reads
- Make listSessions, resolveAgentSession, getAgentHistoryPage synchronous
- Remove unused toBrowserOSSession, filterHttpSessionHistoryMessages,
  normalizeHttpHistoryMessages helpers
- Update route handlers to drop unnecessary async/await
- Update tests to use temp JSONL files instead of mocked HTTP/CLI clients

* fix: restore async route handlers for test compatibility with mocked service

* fix: address review feedback — path traversal guard, lazy reader, exists flag

- Add safePath() to OpenClawJsonlReader that validates resolved paths stay
  within stateRoot, preventing path traversal via crafted agentId values
- Use lazy initialization for jsonlReader (nulled on rebuildRuntimeClients)
  instead of creating a new instance per property access
- Return exists: false from resolveSpecificAgentSession when no session
  matches instead of fabricating a ghost session with sessionId: ''

* feat: add dashboard API and enrich home page agent cards

Server:
- Add summarizeToolActivity() that converts tool events into natural
  language descriptions ("Browsed 3 pages, took 2 screenshots")
- Add getDashboard() to OpenClawService that aggregates per-agent stats
  from JSONL: latest message, activity summary, cost, session count
- Add GET /claw/dashboard endpoint

Client:
- Add useAgentDashboard() React Query hook (10s refetch, 5s stale)
- Rewrite useAgentCardData from async IndexedDB hook to pure
  buildAgentCardData() function merging agent entries with dashboard data
- Add activity summary and cost to AgentCardExpanded footer
- Add activitySummary and costUsd fields to AgentCardData type
- Remove IndexedDB dependency from the home page

* feat: add OpenClawObserver for real-time per-agent status via gateway WS

- Add OpenClawObserver that connects to the OpenClaw gateway WebSocket
  control plane and subscribes to chat broadcast events
- Track per-agent status in real time: working (streaming), idle (turn
  complete), error (run failed), with current tool name
- Auto-connect when gateway control plane becomes available, auto-
  reconnect on disconnect with 5s backoff
- Disconnect observer on stop/shutdown
- Wire live status + currentTool into getDashboard() response
- Update client: AgentOverview includes status + currentTool, card shows
  spinning loader + tool name when agent is working
- Status resolution: per-agent WS status takes precedence over gateway-
  level status for working/error states

* feat: add SSE dashboard stream for real-time agent status on home page

Server:
- Add GET /claw/dashboard/stream SSE endpoint that sends an initial
  snapshot then pushes per-agent status events as they arrive from
  the OpenClaw observer
- Add onAgentStatusChange() to OpenClawService exposing the observer's
  listener for the route layer
- Heartbeat every 15s to keep connections alive

Client:
- useAgentDashboard() now subscribes to EventSource at /claw/dashboard/stream
- SSE snapshot event hydrates the React Query cache immediately
- SSE status events patch individual agent status + currentTool in the
  cache without refetching — agent cards update instantly
- Polling fallback raised to 30s since SSE handles real-time

* fix: observer WS handshake — wait for challenge before sending connect

The OpenClaw gateway sends a connect.challenge event before accepting
the connect request. The observer was sending the connect request on
ws.open which raced with the challenge. Now waits for the challenge
event before sending the handshake.

Also add dangerouslyDisableDeviceAuth to the gateway setup config
batch so the observer can connect without device identity on new
installs.

* fix: JSONL reader falls back to most recent file when sessions.json is stale

OpenClaw's sessions.json can record a Pi session ID that doesn't match
the actual JSONL filename on disk. This happens after context compaction
or session restart — the JSONL file gets a new UUID but sessions.json
keeps the old one.

Previously this caused history to silently disappear (the reader tried
to open a non-existent file and returned empty). Now resolveJsonlPath()
checks if the mapped file exists and, when it doesn't, scans the
sessions directory for the most recently modified .jsonl file as a
fallback.

* feat: add ClawSession state machine for reliable per-agent status

The OpenClawObserver only knows about status changes it witnesses via
WS events. If an agent was already running when the observer connected,
or after a reconnect, statuses were stuck at "unknown".

ClawSession is an in-memory state machine that solves this:

1. Seeds from JSONL on first control plane call — reads the latest
   events for each agent and infers working/idle. A session is "working"
   if the last event is a user.message with no subsequent agent.message,
   or an agent.tool_use with no matching agent.tool_result.

2. Receives live transitions from the WS observer — the observer now
   delegates all state management to ClawSession instead of maintaining
   its own status map.

3. Applies a 5-minute staleness threshold — if the last JSONL event
   is older than 5 minutes, assume idle (handles agent crashes).

Consumers (SSE stream, dashboard endpoint) read from ClawSession and
get correct state from the first call — no "unknown" period.

* fix: remove staleTime so dashboard refetches on every mount

* fix: reset stale working status on WS disconnect, eliminate redundant JSONL reads

- Observer resets all "working" agents to "unknown" when the WS closes,
  preventing agents from appearing stuck as Working indefinitely after
  a gateway restart. ClawSession re-seeds correct state on reconnect.

- getDashboard() now derives latestAgentMessage and cost from the
  already-loaded events array for the latest session instead of calling
  latestAgentMessage() and getSessionStats() which each re-read the
  same JSONL file. Reduces file reads from 3x to 1x per agent.
2026-04-25 19:03:03 +05:30
Neel Gupta
4284e88625 feat: Implement lazy LLM judge for passive monitoring (#777)
* fix: double close on stream controller

* feat: initial lazy llm judge impl

* feat: added regex-based matching to insert button context

* fix: tests & bugfix

fix: redundant truthiness check

* fix(tests): stabilize server suites on dev
2026-04-25 12:52:41 +01:00
Nikhil
0b91c735ab chore: bump server version, offset and patch for release (#814) 2026-04-24 12:05:47 -07:00
Nikhil
d189b50b03 fix: package bundled Lima guest agent (#813)
* fix(build): upload Lima runtime files

* fix(build): stage Lima prefix resources

* fix(vm): resolve bundled Lima prefix

* docs(build): document Lima runtime packaging

* chore: self-review fixes

* fix: address review feedback for PR #813
2026-04-24 12:03:26 -07:00
Nikhil
a407e48209 Prefetch runtime VM cache (#811)
* feat: add runtime vm cache sync

* feat: configure runtime vm cache sync

* feat: prefetch vm cache on startup

* feat: await vm cache before vm startup

* fix: recheck vm cache after prefetch wait

* fix: address vm cache review feedback

* build(server): require VM cache manifest env
2026-04-24 10:41:20 -07:00
shivammittal274
1f75b91fba feat(openclaw): add Claude CLI as a CLI-backed provider (#791)
* feat(openclaw): add Claude CLI as a CLI-backed provider

Extensible registry of "OpenClaw CLI-backed providers" — tools that run
as subprocesses inside the gateway container rather than via an API key.
Claude CLI is the first entry; Gemini CLI / Codex CLI / etc. are
one-line additions in the same shape.

Backend:
- New openclaw-cli-providers/ module: types, registry, claude-cli entry.
- OpenClawService: generic ensureAllCliProvidersInstalled() (runs on
  setup/start/restart/auto-start) and getCliProviderAuthStatus(provider).
- Provider dispatch: resolveProviderForAgent() short-circuits CLI
  providers (no env var, no custom-provider merge) before falling
  through to the API-key resolver. No changes to openclaw-provider-map.
- Container runtime: PATH + NPM_CONFIG_PREFIX env so tools installed
  under /home/node/.npm-global/bin (mounted) are discoverable by
  OpenClaw's child-process spawns and persist across restarts.
- New route: GET /claw/providers/:providerId/auth-status returns
  installed / loggedIn / account / plan / error.

Frontend:
- New openclaw-cli-providers.tsx: mirrors backend registry (id, models,
  authLoginCommand), useOpenClawCliProviderAuthStatus hook (2-s poll
  while enabled), OpenClawCliProviderStatusPanel component.
- AgentsPage: synthesized CLI-provider options merged into the Create
  Agent dropdown, inline status panel, auth modal mounting the existing
  AgentTerminal with provider.authLoginCommand, auto-close on loggedIn.
- AgentTerminal: new optional initialCommand + onSessionExit props
  (ref-based so parent re-renders don't rebuild the PTY).

No global ProviderType changes. No custom container image — runtime
install into the mounted home dir persists across restarts.

* fix(openclaw): address review comments for claude-cli provider

- Drop redundant providerId field from OpenClawCliProviderOption (type
  already carries the same value).
- Reuse SetupInput type in resolveProviderForAgent instead of inlining.
- Split ensureCliProviderInstalled into probe + install so logs
  distinguish "already present" from "freshly installed".
- Narrow union in handleCreate via explicit LlmProviderConfig cast; the
  'in'-based narrowing stopped working once the two option shapes
  overlapped on required fields.

* fix: green up server-api tests after claude-cli additions

- Update container-runtime.test.ts snapshot to include the new
  PATH + NPM_CONFIG_PREFIX env args.
- Add a defensive guard in ensureAllCliProvidersInstalled so test
  mocks that swap runtime for a partial stub without execInContainer
  simply skip the install step; production runtime always provides it.

No production behavior change.

* fix(openclaw): use claude /login for auth flow and render terminal full-page

`claude auth login` in 2.1.x silently discards stdin, so the pasted OAuth
code never reaches claude. Switch to the REPL's `/login` slash command,
which does accept a pasted token. Also render the auth terminal
full-page instead of inside a Radix Dialog — the focus trap was hiding
keyboard events from xterm's helper textarea. Finally, guard the async
WebSocket in AgentTerminal against React 18 StrictMode's double-invoke
so the first mount's orphaned WS doesn't leak a second live session.

- terminal-session: pass PATH on podman exec so user-installed CLIs
  resolve in interactive sessions without manual re-exports.
- claude-cli parseAuthStatus: treat exit-code-1 as a valid "not logged
  in" JSON payload instead of a hard error.

* fix(openclaw): drop unnecessary PATH override on podman exec

`podman exec` inherits the container's run-time env (PATH includes
/home/node/.npm-global/bin via `podman run -e PATH=…`), so the extra
`-e PATH` on the exec call was redundant. Reverts the export of
GATEWAY_PATH and the exec flag added in the previous commit.

* feat(openclaw): show CLI-backed providers in Set Up dialog

The Set Up OpenClaw dialog previously listed only API-key LLM
providers. Add the CLI-backed ones (currently just Claude CLI) so
users can bootstrap the gateway with a Claude.ai-subscription-backed
agent without round-tripping through the Create Agent flow first.

When the user picks a CLI provider at setup, skip the apiKey/baseUrl
fields and open the auth terminal immediately after the gateway comes
up, so /login runs in one click.

* fix(openclaw): robust claude auth-status parsing and cleaner CLI UX

parseClaudeAuthStatus was doing JSON.parse on the entire stdout, which
fails when Lima/nerdctl appends a stderr line like `level=fatal
msg="exec failed with exit code 1"` whenever the inner command exits
non-zero (claude auth status exits 1 when not logged in). The panel
then surfaced the raw output as an error. Switch to a line-by-line
scan that picks the first parseable JSON object — handles trailing
noise and nested JSON fields cleanly.

UI polish around the Setup dialog:
- Hide the "uses your API key" hint when the selected provider is
  CLI-backed — it is inaccurate and confusing.
- When a CLI provider is picked in Setup, show a short helper line
  instead of the status panel (the /auth-status poll would be
  pre-gateway and would always fail). Set Up & Start boots the
  gateway and then auto-opens the auth terminal in one click.
- Track the active CLI provider across both Setup and Create dialogs
  so the auth terminal opens for the right provider regardless of
  which dialog triggered it.

* feat(terminal): make selection + copy work under TUI mouse tracking

Interactive TUIs like `claude /login` enable xterm mouse-tracking,
which forwards every click to the app and disables click-drag text
selection. Our terminal had no escape hatch, so users couldn't grab
the OAuth URL.

Three general-purpose fixes (none CLI-specific):
- macOptionClickForcesSelection: Opt+drag always selects on Mac,
  regardless of what the running program does with mouse events.
- Cmd/Ctrl+A and Cmd/Ctrl+C custom key handler: select-all and copy
  to clipboard via navigator.clipboard, even when the TUI would
  swallow the keys.
- Copy button in the terminal header: writes the current selection
  to the clipboard, or the full visible viewport if nothing is
  selected. One-click escape hatch that works in every state.

Applies to any interactive CLI in our terminal (sudo, vim, claude,
gh auth, etc.), not just the claude login flow.

* fix(terminal): make xterm selection actually visible

Selection was registering internally (xterm-selection layer had
correct width/height rects), but the rectangles rendered in
rgb(252,252,251) — practically invisible against the white
background — so users concluded selection was broken.

Root cause: the theme derived selectionBackground from
`withAlpha(resolveCssColor('--accent-orange'), 0.2)`. When the CSS
var failed to resolve it fell back near-white, and the alpha
compositing against the page background made the result
indistinguishable from the background.

Switch to solid terminal-standard selection colors (VSCode-like
light-blue / dark-indigo). Also set selectionInactiveBackground so
the selection persists when focus moves away (useful while copying).
Drop the now-unused withAlpha helper.

* fix(openclaw): handle pretty-printed JSON in claude auth status parser

claude auth status --json emits multi-line pretty-printed JSON. The previous line-by-line parser never matched, so the UI treated every response as an error and surfaced the raw JSON — even when loggedIn was true. Replace with a brace-matching JSON extractor (string- and escape-aware) that tolerates multi-line JSON, leading banners, trailing lima/nerdctl stderr, and nested objects.

* refactor(openclaw): separate exec streams, argv installs, cleaner async cleanup

Audit-driven cleanup. Net -42 lines, four concrete issues fixed:

1. ContainerRuntime.runInContainer() exposes {exitCode, stdout, stderr}
   from the nerdctl exec (ContainerCli.runCommand already tracked them
   separately; we were just throwing stderr into the same string). The
   40-line hand-rolled brace-matching JSON extractor in claude-cli.ts
   existed only because the prior merged-stream output had lima/
   nerdctl's 'level=fatal' line fused with claude's JSON. parser is
   now JSON.parse(stdout.trim()).

2. Replace shell-based 'sh -lc "npm install -g ${pkg}@latest"' with
   argv: execInContainer(['npm','install','-g','${pkg}@${version}']).
   Registry values no longer flow through a shell (removes injection
   surface from future CLI providers). Pinned version instead of
   @latest (adds npmPackageVersion to the provider type).

3. AgentTerminal: replace the 'let cancelled' + out-of-effect
   disposeSocketBindings pattern with an AbortController scoped to
   the effect and a cleanups[] array. Matches the canonical React 18
   async-effect pattern — no partial-cleanup race if StrictMode
   unmounts between the async await and the resolve.

4. AgentTerminal: drop the full-buffer fallback in the Copy button
   (was copying all 8000 scrollback lines when nothing selected —
   surprising). Button now only copies the actual xterm selection,
   or no-ops silently. Users who want everything can Cmd+A first.
2026-04-24 20:13:18 +05:30
Dani Akash
752f42d1fe refactor: migrate chat history to direct JSONL file reads via Lima filesystem (#808)
* feat: draft agent chat ui exploration

* feat: refine agent chat ui draft

* feat: remove outer frame from agent chat workspace

* fix: offset agent chat for app sidebar

* fix: simplify agent conversation shell

* fix: remove redundant chat header actions

* fix: unify agent conversation headers

* fix: tighten agent chat spacing

* fix: bound agent chat composer height

* fix: remove agent chat page inset

* fix: align agent header height with sidepanel

* fix: center agent composer resting state

* fix: anchor multiline composer controls

* fix: remove focus grid from agent home

* fix: remove redundant agent home header

* fix: constrain home agent composer

* fix: match home composer default posture

* feat: add openclaw chat history APIs

* feat: add claw chat history hydration

* fix: stabilize claw chat viewport layout

* fix: use conversation scroll base for claw chat

* refactor: split claw chat controller responsibilities

* fix: keep active agent turns in memory

* fix: normalize openclaw chat sessions

* refactor: use HTTP client for agent history instead of CLI client

Replace the CLI-based getChatHistory() call in getAgentHistoryPage()
with the HTTP client's getSessionHistory() from PR #795. This uses
the direct HTTP transport to OpenClaw's /sessions/<key>/history
endpoint instead of shelling out through the CLI.

- Add filterHttpSessionHistoryMessages() for flat-string content format
- Add normalizeHttpHistoryMessages() for OpenClawSessionHistoryMessage shape
- Update getAgentHistoryPage() to call getSessionHistory() via httpClient
- Remove unused getChatHistory(), filterOpenClawSystemMessages(),
  normalizeChatHistoryMessages(), and getTextContent()
- Update test mocks from cliClient.getChatHistory to httpClient.getSessionHistory
- Update MutableOpenClawService type: chatClient -> httpClient

* fix: fetch all session messages by iterating OpenClaw pagination

OpenClaw's HTTP history endpoint returns a limited page by default.
When called without a limit, only the first ~27 messages were returned,
causing all newer conversation messages to be silently dropped.

Add fetchAllSessionMessages() that iterates through OpenClaw's cursor-
based pagination (200 messages per page) until hasMore is false, then
feeds the complete message list into the existing BrowserOS normalization
and in-memory pagination layer.

* refactor: migrate chat history from HTTP gateway to direct JSONL file reads

Replace the HTTP-based chat history pipeline (BrowserOS server → OpenClaw
gateway /sessions/:key/history pagination loop) with direct JSONL file reads
from the host filesystem via Lima's virtiofs mount.

- Add OpenClawJsonlReader that reads session JSONL files directly from
  ~/.browseros/vm/openclaw/.openclaw/agents/<id>/sessions/
- Replace fetchAllSessionMessages() HTTP pagination with single file read
- Replace CLI-based listSessions() with sessions.json file reads
- Make listSessions, resolveAgentSession, getAgentHistoryPage synchronous
- Remove unused toBrowserOSSession, filterHttpSessionHistoryMessages,
  normalizeHttpHistoryMessages helpers
- Update route handlers to drop unnecessary async/await
- Update tests to use temp JSONL files instead of mocked HTTP/CLI clients

* fix: restore async route handlers for test compatibility with mocked service

* fix: address review feedback — path traversal guard, lazy reader, exists flag

- Add safePath() to OpenClawJsonlReader that validates resolved paths stay
  within stateRoot, preventing path traversal via crafted agentId values
- Use lazy initialization for jsonlReader (nulled on rebuildRuntimeClients)
  instead of creating a new instance per property access
- Return exists: false from resolveSpecificAgentSession when no session
  matches instead of fabricating a ghost session with sessionId: ''
2026-04-24 13:19:46 +05:30
Nikhil
2f8e36546f fix: resize BrowserOS VM resources (#807) 2026-04-23 18:24:49 -07:00
Nikhil
461dcd29e8 fix: upload Lima resources under vendor prefix (#805) 2026-04-23 17:19:45 -07:00
Nikhil
c6c902a4ab feat: improve dev watch Lima preflights (#802)
* feat: improve dev watch lima preflights

* fix: note vm cache sync duration

* fix: address review feedback for PR #802
2026-04-23 17:16:50 -07:00
Nikhil
6e37742a5a feat: reuse agent command chat for agents page (#803) 2026-04-23 17:09:49 -07:00
Nikhil
1186c2c0d7 merge: feat/new-lima-vm
feat: new vm integration
2026-04-23 16:41:14 -07:00
Nikhil
0288cc040d feat: use rootless nerdctl in BrowserOS VM (#800)
* feat: use rootless nerdctl in BrowserOS VM

* fix: validate openclaw gateway auth before reuse

* fix: forward rootless containerd socket

* fix: address VM review comments
2026-04-23 16:36:51 -07:00
Nikhil
07b7bf5977 feat(build-tools): seed dev agent tarballs (#799)
* feat(build-tools): seed dev agent tarballs

* fix: address review comments for 0423-build_agent_tarball_dev_sync

* chore(build-tools): remove dev cache sync alias
2026-04-23 15:47:00 -07:00
Nikhil
d1a3d67e29 chore(dev): add VM cache setup flow (#798) 2026-04-23 15:47:00 -07:00
Nikhil
35134518f0 fix(vm): use system nerdctl in Lima runtime (#797) 2026-04-23 15:47:00 -07:00
Nikhil Sonti
4083155e81 feat(container): migrate container runtime to nerdctl over Lima VM
Replace the podman-based runtime with nerdctl running inside the Lima
VM introduced in the previous commit. OpenClaw is cut over to the new
VM-backed container runtime; legacy podman code paths are removed.

- New container CLI (lib/container): nerdctl ContainerCli, ImageLoader
  with cache-tarball fallback, shared types
- OpenClaw: container-runtime-factory orchestrates VM lifecycle + gateway
  startup; container-runtime.ts rewritten to speak nerdctl; Linux test
  startup kept disabled behind the factory
- Terminal: session + routes moved onto Lima shell transport; server
  wires the VM-backed runtime via main.ts
- Agent UI: simplify AgentsPage/useOpenClaw after route consolidation
- Remove podman-runtime, podman-overrides, and their tests
- Tests: container-cli, image-loader, container-runtime-factory, and
  updated openclaw/terminal/main suites
2026-04-23 15:46:50 -07:00
Nikhil Sonti
72ef4f068e feat(vm): add Lima-based BrowserOS VM runtime
Introduce a new VM runtime layer using Lima for running containerised
workloads on macOS. Lifecycle covers decompress/create/start/stop with
stubs for upgrade/reset plus version-mismatch warnings.

- Foundation modules: paths, errors, manifest, telemetry
- lima.yaml generator + typed limactl wrapper with structured debug logging
- ssh ControlMaster transport for fast in-VM commands
- Ubuntu 24.04 minimal template, containerd default, 30GiB overlay disk
- browseros-dir helpers (getLimaHomeDir, getVmStateDir, getVmDisksDir);
  OpenClaw dir moves into VM state dir
- Test helpers (fake-limactl, fake-ssh, test-env), vm-smoke integration
  coverage, NODE_ENV propagation for spawned server test groups
2026-04-23 15:46:25 -07:00
Nikhil
6b6ed1582c feat(openclaw): HTTP session history endpoint (JSON + SSE) (#795)
* refactor(openclaw): rename http chat client to http client

Session history is about to land on the same HTTP client. 'Chat client'
will no longer describe it, so rename the class, file, and service field
up front. No behavior change.

* feat(openclaw): add session history fetch + sse stream to http client

Adds getSessionHistory (JSON) and streamSessionHistory (SSE) to the
OpenClaw HTTP client. Both target GET /sessions/<key>/history on the
loopback gateway, reusing the same bearer-token auth as streamChat.

- 404 from the gateway surfaces as OpenClawSessionNotFoundError so
  callers can map it to a typed HTTP status.
- The SSE path parses named 'history', 'message', and 'error' events
  into a typed OpenClawSessionHistoryEvent union.
- AbortSignal propagates to fetch and cancels the reader mid-stream.

* feat(openclaw): expose session history over GET /claw/session/:key/history

Wire the new getSessionHistory / streamSessionHistory service methods
through a route that defaults to JSON and upgrades to SSE when the
client sends Accept: text/event-stream.

- OpenClawSessionNotFoundError lives in errors.ts alongside the other
  OpenClaw errors so routes can import it from one place.
- The route propagates c.req.raw.signal into streamSessionHistory so
  client disconnects cancel the upstream fetch.
- Route tests cover the JSON path (with query param forwarding), the
  404 path, and the SSE framing.

* chore(openclaw): drop NaN from session history route limit param
2026-04-23 11:19:16 -07:00
Nikhil
a3764e7599 feat(build-tools): add cache:sync:dev for local tarball seeding (#794)
Seeds ~/.browseros-dev/cache/vm/ from ./dist/ without touching R2, so
devs can test the server against a freshly-built tarball before anything
is published to cdn.browseros.com. Hardcodes arm64 since all devs are on
Apple Silicon; refuses to run unless NODE_ENV=development; idempotent
(skips copy on sha256 match).

Also fixes the R2_BUCKET default in .env.sample from browseros-artifacts
to browseros to match the actual bucket.
2026-04-23 10:33:51 -07:00
Nikhil
c656f6236c feat: ship Lima template for BrowserOS VM (#787)
* feat(build-tools): add Lima template for BrowserOS VM

* feat(build-tools): remove build-disk pipeline and recipe directory

Task 2 verification removed the scripts, recipe directory, workflow, and package scripts. Typecheck remains green here because manifest disk fields are removed in the next task, so the plan's expected missing-import failure does not apply yet.

* feat(build-tools): rename VmManifest to AgentManifest, drop disk fields

* feat(build): stage Lima template into server resources

Verified local-resource staging with: bun scripts/build/server.ts --target=darwin-arm64 --ci. The template was copied to dist/prod/server/darwin-arm64/resources/vm/browseros-vm.yaml and included in the zip. bun run build:server:test still fails on the pre-existing R2 limactl resource with: The specified key does not exist.

* docs(build-tools): Lima template dev loop + record D9

Updated the build-tools README in this worktree. Also recorded D9 in the canonical external spec file at /Users/shadowfax/llm/code/browseros-project/grove-ref/browseros-main/specs/decisions.md, which is outside this git checkout.

* chore(build-tools): sweep orphaned references to retired disk pipeline

* chore: self-review fixes
2026-04-22 17:17:12 -07:00
Nikhil
4d660874ad feat: consolidate build tools package (#785)
* feat(build-tools): scaffold package + cache dir helpers

* feat(build-tools): manifest types + R2 helper

* feat(build-tools): build-disk script with virt-customize + zstd

* feat(build-tools): build-tarball script

* feat(build-tools): emit-manifest + cache:sync

* ci(build-tools): independent build-vm + build-agent workflows

* chore: remove legacy container packages + workflows

* fix: address review feedback for PR #785

* fix: stabilize VM build DNS in CI

* fix: prioritize arm64 build workflows

* fix: keep arm64 VM recipe simple

* fix: set VM build DNS in apt command

* fix: avoid guest DNS for VM package install

* fix: limit VM PR checks to build-tools validation
2026-04-22 16:23:11 -07:00
Nikhil
819887a2c5 feat(vm-container): WS1 VM disk image pipeline (#783)
* feat(vm-container): ship the WS1 VM disk image pipeline

New Bun/TS workspace package @browseros/vm-container that produces a
reproducible, versioned Debian 12 + Podman qcow2 disk image for arm64 and
x64, and publishes it to Cloudflare R2 under vm/<version>/ with a per-
version manifest.json and a latest.json pointer.

- virt-customize-driven build with a git-tracked recipe DSL.
- zstd-compressed artifacts; sha256 sidecars for compressed + uncompressed.
- Public surface at @browseros/vm-container/schema exposes zod-validated
  VmManifest + R2 key helpers for WS4 to import; /download is a stub
  landing pad for WS4 to fill in.
- Rollback on partial upload failure: any exception after the first
  successful put deletes all previously uploaded keys for that version.
- GHA workflow build-vm-container.yml runs a matrix build per arch on
  native runners, an x64 Lima boot smoke test, and a gated publish job.
- Full unit coverage for arch, r2-keys, manifest, recipe parser, and
  publish (rollback + happy path via aws-sdk-client-mock).

* fix(vm-container): address review comments

- Split buildDisk into prepareCustomizedDisk + finalizeArtifacts for
  testability.
- Replace resolvePinnedSha's sentinel-prefix check with a positive
  sha256-hex regex test, switch base-image.ts placeholder to empty string.
- Drop unused R2_VM_PREFIX from .env.example; document CDN_BASE_URL
  override precedence in README.
- Replace SSH host-key explicit list in recipe with `ssh_host_*` glob so
  .pub keys and future key types are also removed.
- lima-boot: introduce BunRequestInit type for the unix fetch option and
  reject empty limactlPath loudly.
- Extend publish test suite: mid-manifest-upload failure path verifies
  both arches' qcow+sha are rolled back and latest.json is never written.
- Add missing tests: parseArch('ARM64') case-sensitivity rejection,
  composeVirtCustomizeArgv unresolved-substitution pass-through.

* fix(vm-container): pin a real Debian snapshot, switch verify to SHA-512, streaming download

- Pin Debian base to bookworm/20260413-2447 with real SHA-512 values
  from upstream SHA512SUMS (the sentinel placeholder never corresponded
  to a real build). Debian cloud images only publish SHA512SUMS today,
  so switch base-image verification to SHA-512 throughout: rename
  BaseImage.sha256 → sha512, manifest field base_image_sha256 →
  base_image_sha512, base_image.sha256_url → sha512_url,
  debianSha256SumsUrl → debianSha512SumsUrl. Our own artifact hashes
  (compressed_sha256, uncompressed_sha256, recipe_sha256) stay SHA-256.
- Fix downloadTo: previous Bun.write(dest, response) buffered the
  entire 300 MB response before writing (100% CPU, empty dir). Replace
  with a getReader() loop that streams chunks through Bun.file().writer().
- build CLI now auto-derives --version from today's date when omitted
  (defaults to YYYY.MM.DD-dev1); explicit --version still overrides.
  Broaden CALVER_REGEX to accept alphanumeric suffixes so -dev1/-rc1
  tags are valid. New todayCalver() helper.
- Update GHA workflow fallback to github.run_number (shorter) instead
  of run_id.

* fix(vm-container): resolve copy-in paths against recipeDir after substitution

The copy-in path resolver checked op.src.startsWith('/') before running
the {placeholder} substitution, so an absolute-after-substitution path
like {manifest_tmp} → /tmp/vm-dist/manifest-stub-arm64.json was treated
as relative and joined against recipeDir, producing a nonexistent path.
Check the *substituted* value for absoluteness via path.isAbsolute.

* fix: address review comments for 0422-ws1_vm_disk_pipeline

* fix(ci): repair vm-container workflow

* fix(ci): expose vm build logs on failure

* fix(vm-container): expose base_image_sha256 in manifest per PRD

The published manifest contract (consumed by WS4) now uses base_image_sha256
as the PRD specified. Internally the build still verifies the downloaded
Debian base against the pinned sha512 (that's what Debian actually signs in
SHA512SUMS) — then hashes the same bytes as sha256 and records that in the
manifest. One extra digest pass of a ~300 MB file; negligible.

- manifest.json: base_image_sha256 replaces base_image_sha512; sha512_url
  removed (not needed — sha256 is the consumer-facing hash).
- CLI: --base-image-sha256 override validates against the locally-computed
  sha256 after download.
- BuildResult.baseImage gains sha256 alongside sha512.
- Tests updated to the new field.

The auth.json bug (reviewer #2) is resolved: the source file is
recipe/auth.json and the recipe emits `copy-in auth.json:/etc/containers/`
so libguestfs writes /etc/containers/auth.json.

* ci(vm-container): fix supermin kernel-read + rename sha512 inputs to sha256

- Ubuntu 24.04 GHA runners ship /boot/vmlinuz-* as mode 0600, which blocks
  libguestfs's supermin appliance builder when virt-customize runs as a
  non-root user. Chmod 0644 before the build — canonical CI workaround.
- Rename workflow_dispatch input base_image_sha512 → base_image_sha256
  and CLI flag --base-image-sha512 → --base-image-sha256 to match the
  orchestrator's renamed override.

* ci(vm-container): give runner KVM access + install passt for libguestfs

The supermin fix got us past appliance-build, but virt-customize then hit
"passt exited with status 1". The passt networking helper misbehaves when
libguestfs falls back to TCG emulation, which happens because the runner
user isn't in the kvm group even though /dev/kvm exists on the GHA host.

- chmod 0666 /dev/kvm → libguestfs uses hardware acceleration, avoids TCG.
- install passt explicitly so the networking helper is present and current.

* ci(vm-container): disable passt to force libguestfs slirp fallback

libguestfs 1.54+ prefers passt for guest networking, but the passt binary
on GHA ubuntu-24.04 exits with status 1 when invoked from the appliance
— an AppArmor/capability issue that doesn't surface a useful diagnostic.
The reliable workaround is to remove passt so libguestfs picks QEMU's
built-in user-mode SLIRP as the network backend. SLIRP is slower but
functional and doesn't require escalated privileges.
2026-04-22 14:04:00 -07:00
Nikhil
114d5e3a9f feat: add agent container tarball pipeline (#782)
* feat: add agent container tarball pipeline

* docs: add agent-container env sample

* refactor: simplify agent container pipeline

* fix: address review feedback for PR #782

* fix: emit clean matrix JSON in CI

* fix: align agent container artifact paths
2026-04-22 13:14:27 -07:00
Nikhil
ecba7de221 fix: address PR review comments for 0422-ws3_lima_resources (#780)
- Guard uploaded_keys append with !dry_run so the rollback list
  never contains keys for objects that were never written.
- Prefer GITHUB_ACTOR over local OS username for manifest.uploaded_by;
  manifest.json is CDN-fronted so leaking a developer's login is
  unnecessary (falls back to 'local').
- Extend test_windows_has_no_stale_third_party to cover bun.exe/rg.exe
  too, matching the macOS forbidden-set pattern.
2026-04-22 10:47:25 -07:00
Nikhil
123a13fe62 feat(build): swap podman server resources for Lima (WS3) (#778)
* feat(build): swap podman server resources for Lima (WS3)

- Upload limactl (arm64 + x64) to R2 via new 'browseros upload lima' CLI.
- Rewrite scripts/build/config/server-prod-resources.json: 2 Lima entries,
  12 podman-family entries removed.
- Update codesign metadata (server_binaries.py) to add limactl, drop podman
  family. Sign modules need no edits (data-driven).
- Delete orphaned podman-{vfkit,krunkit} entitlement plists.
- Release-gating note in browseros-agent/CLAUDE.md: don't cut releases off
  dev between this commit and WS6 landing (OpenClaw still invokes podman).

* fix: address review comments for 0422-ws3_lima_resources

- Tighten _find_limactl_member to match exactly .../bin/limactl via
  Path.parts, avoiding incidental matches like 'xbin/limactl'.
- Fall back USER -> USERNAME -> 'unknown' for uploaded_by so Windows
  shells don't all record 'unknown'.
- Comment the broad except in upload_lima to explain why rollback
  must fire for any mid-loop failure.

* chore: drop bun + rg from Windows sign list

These executables are already absent from server-prod-resources.json (no
Windows entries shipped); keeping them in the sign list produces
"Binary not found" warnings on every Windows build.
2026-04-22 10:40:53 -07:00
Nikhil
5ccdbaf87f feat(openclaw): lifecycle progress banner + live podman readiness (#772)
* fix(openclaw): serialize lifecycle operations

* feat(openclaw): lifecycle progress banner and live podman readiness check

* fix: address review comments for openclaw-lifecycle-progress
2026-04-21 07:59:33 -07:00
Nikhil
0650f21c80 fix(openclaw): allocate gateway host port dynamically + name the two ports distinctly (#771)
* feat(openclaw): dynamically allocate and persist gateway host port

The gateway container always listens on OPENCLAW_GATEWAY_CONTAINER_PORT
(18789) internally, but that port may be taken on the user's host. Allocate
a free host port on each lifecycle transition, persist it to
~/.browseros/openclaw/.openclaw/runtime-state.json, and prefer the
persisted value on subsequent starts so the mapping is stable.

Split the naming so the two sides of the -p mapping are no longer
ambiguous: the shared constant becomes OPENCLAW_GATEWAY_CONTAINER_PORT
and the service/spec/chat-client/runtime probes all use hostPort for
the mapped host-side port.

* fix(openclaw): remove duplicate Podman overrides card from status panels
2026-04-20 17:32:10 -07:00
Dani Akash
e80ec467f4 feat: wire lazy monitoring to OpenClaw chat handoff (#768)
* feat: add lazy monitoring substrate

* feat: wire lazy monitoring to openclaw chat handoff

* test: cover openclaw chat history handoff

* fix: reject concurrent monitored chats
2026-04-20 21:52:03 +05:30
Dani Akash
41374439c4 feat: add passive lazy monitoring substrate for MCP tool calls (#766)
* feat: add lazy monitoring substrate

* fix: validate monitoring run ids

* fix: harden monitoring storage recovery
2026-04-20 21:10:09 +05:30
Dani Akash
ad99cd6cc1 fix: restore openai-compatible OpenClaw providers (#767)
* fix(openclaw): restore openai-compatible providers

* fix(openclaw): preserve custom provider model lists
2026-04-20 20:25:37 +05:30
Nikhil
47fc9e1292 feat(openclaw): user-supplied Podman binary path override (#759)
* feat(openclaw): user-supplied Podman binary path override

Expose the existing `configurePodmanRuntime({ podmanPath })` knob as a UI
input on the Agents page so users blocked by the bundled gvproxy helper
discovery bug can install their own Podman (e.g. `brew install podman`)
and point BrowserOS at it.

- podman-overrides.ts: persist {podmanPath} at ~/.browseros/.openclaw/
- openclaw-service: applyPodmanOverrides/getPodmanOverrides, rebuilds
  ContainerRuntime + CLI clients in place (no server restart needed)
- routes: GET/POST /claw/podman-overrides with absolute-path + existsSync
  validation
- main: load override on boot, pass resourcesDir into the service so
  clearing the override restores bundled fallback
- AgentsPage: PodmanOverridesCard rendered inline in the degraded /
  uninitialized / error cards and as a collapsible standalone section

Dev mode is unchanged; prod gets the same lever dev has had all along.

* refactor(openclaw): address review comments for podman-path override

- extract getPodmanOverrideValidationError() to mirror the existing
  getCreateAgentValidationError() pattern in openclaw.ts
- extract rebuildRuntimeClients() so applyPodmanOverrides doesn't
  re-spell the three-step runtime/CLI-client reinit
- rename shadowing local path -> overridesPath in loadPodmanOverrides

* fix(openclaw): clear gateway log tail before swapping runtime

rebuildRuntimeClients replaces this.runtime but the cached stopLogTail
still closes over the old runtime's log-tail process. The existing
guard in startGatewayLogTail (if (this.stopLogTail) return) would then
short-circuit the next restart and leave the new runtime without a
tail. Clear it inside the helper so the rebuild is self-consistent
regardless of caller order.

* fix(openclaw): check podmanPath executability and note singleton mutation

- validator: after existsSync, accessSync(X_OK) so a non-executable file
  fails fast at save time with a clear 400 instead of a cryptic spawn
  error later. Added a matching route test.
- applyPodmanOverrides: one-line comment flagging the intentional
  module-level PodmanRuntime singleton mutation so future readers know
  this is by design, not an accident.
2026-04-18 17:27:25 -07:00
Nikhil
2a61dcbc58 fix: remove podman compose from OpenClaw runtime (#758)
* refactor: rename OpenClaw runtime away from compose semantics

* feat: run OpenClaw containers with direct podman commands

* test: assert exact podman run args

* fix: stage direct runtime container migration safely

* refactor: switch OpenClaw service to direct podman runtime

* test: cover direct-runtime lifecycle paths in openclaw service

* fix: handle legacy openclaw gateway container during runtime cutover

* chore: remove OpenClaw compose resources from server build

* refactor: drop obsolete setup-command overload

* fix: remove dead OpenClaw runtime env file flow

* fix: restore scoped OpenClaw gateway container name

* test: assert scoped OpenClaw terminal container name

* fix: make OpenClaw gateway removal idempotent

* fix: harden OpenClaw setup container lifecycle
2026-04-18 13:53:18 -07:00
Nikhil
f5a2b7315c fix: run all browseros-agent tests from root (#750)
* fix: run full browseros-agent test suite

* fix: stabilize server test reporting in CI

* fix: address PR review feedback

* refactor: extract server core test runner

* refactor: group server tests by filesystem

* fix: align CI suites with server test groups

* fix: provision server env for all CI suites

* fix: stabilize ci checks

* fix: report real test counts in ci
2026-04-17 17:26:44 -07:00
Nikhil
6de3b3422c fix: package OpenClaw compose resource (#749)
* fix: package openclaw compose resource

* fix: address PR review comments for docker-compose-missing
2026-04-17 15:01:59 -07:00
Nikhil
224b6cd3a8 chore: remove bun and ripgrep prod resources (#748) 2026-04-17 13:03:42 -07:00
Nikhil
7baee8d57e chore: release server alpha - 0.0.88 (#747) 2026-04-17 12:44:41 -07:00
Nikhil
e8e8c36fdb fix: pin OpenClaw image to 2026.4.12 (#746)
* fix: pin OpenClaw image to 2026.4.12

* fix: address PR review comments for 0417-openclaw-image-pin
2026-04-17 12:14:37 -07:00
Nikhil
3810005457 refactor: stabilize local OpenClaw integration (#741)
* feat(openclaw): add CLI client

* fix(openclaw): swap service to cli client

* fix(openclaw): restore mixed json parsing

* fix(openclaw): validate agent list payloads

* fix(openclaw): simplify cli client boundary

* fix(openclaw): simplify cli client boundary

* fix(openclaw): prefer outer config json payloads

* fix(openclaw): ignore trailing config log payloads

* refactor(openclaw): bootstrap config through cli

* fix(openclaw): narrow bootstrap ownership

* fix(openclaw): avoid noop key restarts

* fix(openclaw): enforce supported provider sync

* refactor(openclaw): remove agent role contract

* fix(openclaw): migrate legacy state and apply model updates

* fix(openclaw): migrate legacy agent state

* fix(openclaw): harden state updates

* refactor: stabilize local OpenClaw bootstrap and chat auth

* fix(openclaw): propagate container env and drop legacy paths

Compose now loads provider creds from .openclaw/.env and passes the
gateway token through, so in-container CLI commands (tui, doctor,
config) authenticate correctly and the gateway process sees
OPENROUTER_API_KEY. Service ensures the state env file exists and
rewrites the compose env with the token before composeUp in setup,
start, and tryAutoStart. Podman machine gets larger defaults and the
container enables NODE_COMPILE_CACHE + OPENCLAW_NO_RESPAWN. Legacy
state migration, the unused WebSocket gateway-client, memorySearch,
and thinking defaults are removed.
2026-04-17 11:00:07 -07:00
Nikhil
688f7962cb fix: rerun dev port cleanup before server restarts (#745) 2026-04-17 08:13:22 -07:00
Felarof
526d784d82 chore: add .auctor entries to gitignore (#739)
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-16 18:00:24 -07:00
Nikhil
331fec07e6 fix: use separate BrowserOS dir in development (#736) 2026-04-16 16:42:06 -07:00
Nikhil
0652ee8ca8 feat: better hidden windows (#730)
* feat: better hidden windows

* fix: addressing review comments
2026-04-16 16:33:12 -07:00
Nikhil
156f5dbc5d feat: redesign OpenClaw control plane around CLI and HTTP (#735)
* feat: move OpenClaw control plane to CLI and HTTP

* fix: address PR review comments for 0416-openclaw_cli_http_redesign
2026-04-16 16:29:26 -07:00
Nikhil
ebd3200cfe feat(build): add arm64-only macOS release config (#728)
Introduces release.macos.arm64.yaml for single-architecture arm64
macOS release builds. Mirrors the windows/linux single-arch pattern
(configure -> compile -> sign_macos -> package_macos -> upload),
skipping the universal_build module to avoid the x64 cross-compile
and lipo merge. Reuses the sparkle_setup step and the same
notarization env vars as the universal macOS config.
2026-04-16 13:09:46 -07:00
Nikhil
4172daa130 chore: bump PATCH and OFFSET (#727) 2026-04-16 13:05:01 -07:00
Nikhil
c1b1e53a86 feat(ota): bundle full server resources tree in Sparkle payload (#726)
* feat(ota): bundle full server resources tree (server + third_party bins)

The OTA Sparkle payload now ships the complete resources/ tree the agent
build produced, not just browseros_server. Every third-party binary (bun,
ripgrep, podman, gvproxy, vfkit, krunkit, podman-mac-helper, win-sshproxy)
flows to OTA-updated installs so podman integration works for users on the
OTA channel, matching fresh Chromium-build installs.

Extract the per-binary sign table into build/common/server_binaries.py so
the Chromium-build sign path (modules/sign/) and OTA sign path (modules/ota/)
share a single source of truth. Adding a new third-party dep is now a
one-file edit that both paths pick up automatically; unknown executables
under resources/bin/ are a hard error at release time.

* fix(ota): address review comments on bundle signing flow

- Avoid double-zipping during notarization: add notarize_macos_zip for
  pre-built Sparkle bundles so notarytool submits the zip directly
  instead of re-wrapping it through ditto --keepParent (Apple's service
  does not descend into nested archives). Keep notarize_macos_binary for
  single-binary callers. Share credential setup + submit logic via
  internal helpers.
- Fail fast on unknown executables in sign_server_bundle_macos: collect
  the unknown-files list before any codesign call so a missing shared-
  table entry aborts in seconds, not after a full signing round.
- Drop dead get_entitlements_path helper (no callers remain after the
  bundle refactor).

* fix(ota): address PR review comments (greptile + claude)

- sign_server_bundle_macos filters to executables only (p.is_file() +
  not p.is_symlink() + os.access X_OK) before applying the unknown-file
  guard. Non-Mach-O files (configs, dylibs, etc.) under resources/bin/
  no longer cause misleading 'unknown executable' hard failures.
- sign_server_bundle_windows now hard-errors on a missing expected
  binary instead of silently skipping it. Symmetric with the macOS
  guard — an incomplete bundle must not publish.
- ServerOTAModule.execute() uses tempfile.TemporaryDirectory context
  managers for both the download and staging roots so they are cleaned
  up on every path, including failures.
- Per-platform sign/notarize/Sparkle-sign failures now raise RuntimeError
  instead of silently skipping the platform — a release pipeline can no
  longer omit a target while reporting success.
- Move import os and import shutil to the top of ota/sign_binary.py.
- Drop unused log_error import from ota/server.py.

* chore: bump server
2026-04-16 12:59:49 -07:00
Nikhil
d653883e99 fix(ci): add PR comment with test summary (#724)
* fix(ci): add PR comment with test summary and block on failure

Add a `comment` job to the test workflow that parses JUnit XML artifacts
and posts a sticky PR comment showing pass/fail counts per suite, with
failed test names listed in a collapsible section and a link to the run.

Guards against fork PRs (read-only token) and stale overlapping runs
(skips comment if PR head has moved past our SHA).

* fix(ci): use payload SHA for staleness check, handle missing artifacts

- Replace context.sha (merge commit SHA) with
  context.payload.pull_request.head.sha so the staleness guard
  compares the correct values and the comment actually gets posted
- Add continue-on-error to download-artifact so cancelled runs
  gracefully fall through to the "no test results" message

* fix(ci): show warning icon for zero-test suites instead of failure
2026-04-15 21:35:58 -07:00
272 changed files with 29568 additions and 5284 deletions

157
.github/workflows/build-agent.yml vendored Normal file
View File

@@ -0,0 +1,157 @@
name: build-agent
on:
workflow_dispatch:
inputs:
agent:
description: "Agent name from bundle.json"
required: true
type: string
default: openclaw
publish:
description: "Upload to R2 and merge manifest slice"
required: false
default: false
type: boolean
pull_request:
paths:
- "packages/browseros-agent/packages/build-tools/**"
- ".github/workflows/build-agent.yml"
env:
BUN_VERSION: "1.3.6"
PKG_DIR: packages/browseros-agent/packages/build-tools
permissions:
contents: read
jobs:
check:
runs-on: ubuntu-24.04
steps:
- uses: actions/checkout@v4
- uses: oven-sh/setup-bun@v2
with:
bun-version: ${{ env.BUN_VERSION }}
- working-directory: packages/browseros-agent
run: bun install --frozen-lockfile
- working-directory: packages/browseros-agent
run: bun run --filter @browseros/build-tools typecheck
- working-directory: packages/browseros-agent
run: bun run --filter @browseros/build-tools test
build:
needs: check
strategy:
fail-fast: false
matrix:
include:
- arch: arm64
runner: ubuntu-24.04-arm
runs-on: ${{ matrix.runner }}
steps:
- uses: actions/checkout@v4
- uses: oven-sh/setup-bun@v2
with:
bun-version: ${{ env.BUN_VERSION }}
- name: Install podman
run: |
sudo apt-get update
sudo apt-get install -y podman
- working-directory: packages/browseros-agent
run: bun install --frozen-lockfile
- name: Build tarball
working-directory: ${{ env.PKG_DIR }}
env:
AGENT: ${{ inputs.agent || 'openclaw' }}
OUT: ${{ github.workspace }}/dist/images
run: bun run build:tarball -- --agent "$AGENT" --arch "${{ matrix.arch }}" --output-dir "$OUT"
- uses: actions/upload-artifact@v4
with:
name: tarball-${{ inputs.agent || 'openclaw' }}-${{ matrix.arch }}
path: dist/images/
retention-days: 7
smoke:
needs: build
runs-on: ubuntu-24.04-arm
steps:
- uses: actions/checkout@v4
- uses: oven-sh/setup-bun@v2
with:
bun-version: ${{ env.BUN_VERSION }}
- uses: actions/download-artifact@v4
with:
name: tarball-${{ inputs.agent || 'openclaw' }}-arm64
path: dist/images
- name: Install podman
run: |
sudo apt-get update
sudo apt-get install -y podman
- working-directory: packages/browseros-agent
run: bun install --frozen-lockfile
- name: Smoke test tarball
working-directory: ${{ env.PKG_DIR }}
env:
AGENT: ${{ inputs.agent || 'openclaw' }}
run: |
set -euo pipefail
tarball="$(find "$GITHUB_WORKSPACE/dist/images" -name "${AGENT}-*-arm64.tar.gz" -print -quit)"
if [ -z "$tarball" ]; then
echo "missing arm64 tarball artifact for ${AGENT}" >&2
exit 1
fi
bun run smoke:tarball -- --agent "$AGENT" --arch arm64 --tarball "$tarball"
publish:
needs: [build, smoke]
if: ${{ github.event_name == 'workflow_dispatch' && inputs.publish == true }}
runs-on: ubuntu-24.04
environment: release
concurrency:
group: r2-manifest-publish
cancel-in-progress: false
steps:
- uses: actions/checkout@v4
- uses: oven-sh/setup-bun@v2
with:
bun-version: ${{ env.BUN_VERSION }}
- uses: actions/download-artifact@v4
with:
pattern: tarball-*
path: dist/images
merge-multiple: true
- working-directory: packages/browseros-agent
run: bun install --frozen-lockfile
- name: Upload tarballs to R2
working-directory: ${{ env.PKG_DIR }}
env:
R2_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}
R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
R2_BUCKET: ${{ secrets.R2_BUCKET }}
run: |
set -euo pipefail
for file in "$GITHUB_WORKSPACE"/dist/images/*.tar.gz; do
base="$(basename "$file")"
bun run upload -- --file "$file" --key "vm/images/$base" --content-type "application/gzip" --sidecar-sha
done
- name: Merge agent slice into manifest
working-directory: ${{ env.PKG_DIR }}
env:
AGENT: ${{ inputs.agent || 'openclaw' }}
R2_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}
R2_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
R2_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
R2_BUCKET: ${{ secrets.R2_BUCKET }}
run: |
set -euo pipefail
mkdir -p dist/images
cp -R "$GITHUB_WORKSPACE"/dist/images/* dist/images/
bun run download -- --key vm/manifest.json --out dist/baseline-manifest.json
bun run emit-manifest -- \
--slice "agents:${AGENT}" \
--dist-dir dist \
--merge-from dist/baseline-manifest.json \
--out dist/manifest.json
bun run upload -- --file dist/manifest.json --key vm/manifest.json --content-type "application/json"

View File

@@ -43,6 +43,12 @@ jobs:
working-directory: packages/browseros-agent
run: bun install --ignore-scripts && bun run build:agent-sdk
- name: Install Python eval dependencies
run: pip install agisdk requests
- name: Clone WebArena-Infinity
run: git clone --depth 1 https://github.com/web-arena-x/webarena-infinity.git /tmp/webarena-infinity
- name: Install xvfb
run: sudo apt-get update && sudo apt-get install -y xvfb
@@ -57,9 +63,11 @@ jobs:
working-directory: packages/browseros-agent/apps/eval
env:
FIREWORKS_API_KEY: ${{ secrets.FIREWORKS_API_KEY }}
OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
NOPECHA_API_KEY: ${{ secrets.NOPECHA_API_KEY }}
BROWSEROS_BINARY: /usr/bin/browseros
WEBARENA_INFINITY_DIR: /tmp/webarena-infinity
EVAL_CONFIG: ${{ github.event.inputs.config || 'configs/browseros-agent-weekly.json' }}
run: |
echo "Running eval with config: $EVAL_CONFIG"
@@ -81,6 +89,8 @@ jobs:
- name: Generate trend report
if: success()
timeout-minutes: 5
continue-on-error: true
working-directory: packages/browseros-agent
env:
EVAL_R2_ACCOUNT_ID: ${{ secrets.EVAL_R2_ACCOUNT_ID }}

View File

@@ -30,12 +30,54 @@ jobs:
fail-fast: false
matrix:
include:
- suite: tools
test_path: tests/tools
junit_path: test-results/tools.xml
- suite: integration
test_path: tests/server.integration.test.ts
junit_path: test-results/integration.xml
- suite: server-agent
command: (cd apps/server && bun run test:agent)
junit_path: test-results/server-agent.xml
needs_browser: false
- suite: server-api
command: (cd apps/server && bun run test:api)
junit_path: test-results/server-api.xml
needs_browser: false
- suite: server-skills
command: (cd apps/server && bun run test:skills)
junit_path: test-results/server-skills.xml
needs_browser: false
- suite: server-tools
command: (cd apps/server && bun run test:tools)
junit_path: test-results/server-tools.xml
needs_browser: true
- suite: server-browser
command: (cd apps/server && bun run test:browser)
junit_path: test-results/server-browser.xml
needs_browser: false
- suite: server-integration
command: (cd apps/server && bun run test:integration)
junit_path: test-results/server-integration.xml
needs_browser: true
- suite: server-sdk
command: (cd apps/server && bun run test:sdk)
junit_path: test-results/server-sdk.xml
needs_browser: true
- suite: server-root
command: (cd apps/server && bun run test:root)
junit_path: test-results/server-root.xml
needs_browser: false
- suite: agent
command: bun run test:agent
junit_path: test-results/agent.xml
needs_browser: false
- suite: eval
command: bun run test:eval
junit_path: test-results/eval.xml
needs_browser: false
- suite: agent-sdk
command: bun run test:agent-sdk
junit_path: test-results/agent-sdk.xml
needs_browser: false
- suite: build
command: bun run test:build
junit_path: test-results/build.xml
needs_browser: false
steps:
- name: Checkout code
@@ -48,6 +90,7 @@ jobs:
run: bun ci
- name: Resolve BrowserOS cache key
if: matrix.needs_browser == true
id: browseros-cache-key
run: |
set -euo pipefail
@@ -62,6 +105,7 @@ jobs:
echo "key=browseros-appimage-${{ runner.os }}-$cache_key" >> "$GITHUB_OUTPUT"
- name: Restore BrowserOS cache
if: matrix.needs_browser == true
id: browseros-cache
uses: actions/cache@v4
with:
@@ -69,13 +113,14 @@ jobs:
key: ${{ steps.browseros-cache-key.outputs.key }}
- name: Download BrowserOS
if: steps.browseros-cache.outputs.cache-hit != 'true'
if: matrix.needs_browser == true && steps.browseros-cache.outputs.cache-hit != 'true'
run: |
mkdir -p .ci/bin
curl -fsSL "$BROWSEROS_APPIMAGE_URL" -o .ci/bin/BrowserOS.AppImage
chmod +x .ci/bin/BrowserOS.AppImage
- name: Prepare BrowserOS wrapper
if: matrix.needs_browser == true
run: |
mkdir -p .ci/bin
cat > .ci/bin/browseros <<'EOF'
@@ -96,16 +141,23 @@ jobs:
BROWSEROS_BINARY: ${{ github.workspace }}/packages/browseros-agent/.ci/bin/browseros
BROWSEROS_TEST_HEADLESS: "true"
BROWSEROS_TEST_EXTRA_ARGS: --no-sandbox --disable-dev-shm-usage
BROWSEROS_JUNIT_PATH: ${{ github.workspace }}/packages/browseros-agent/${{ matrix.junit_path }}
run: |
set +e
mkdir -p test-results
cd apps/server
bun run test:cleanup
bun --env-file=.env.development test "${{ matrix.test_path }}" --reporter=junit --reporter-outfile="../../${{ matrix.junit_path }}"
${{ matrix.command }}
exit_code=$?
cd ../..
if [ ! -f "${{ matrix.junit_path }}" ]; then
cat > "${{ matrix.junit_path }}" <<EOF
if [ "$exit_code" = "0" ]; then
cat > "${{ matrix.junit_path }}" <<EOF
<?xml version="1.0" encoding="UTF-8"?>
<testsuites tests="0" failures="0">
<testsuite name="${{ matrix.suite }}" tests="0" failures="0">
</testsuite>
</testsuites>
EOF
else
cat > "${{ matrix.junit_path }}" <<EOF
<?xml version="1.0" encoding="UTF-8"?>
<testsuites tests="1" failures="1">
<testsuite name="${{ matrix.suite }}" tests="1" failures="1">
@@ -115,6 +167,7 @@ jobs:
</testsuite>
</testsuites>
EOF
fi
fi
echo "exit_code=$exit_code" >> "$GITHUB_OUTPUT"
@@ -136,3 +189,124 @@ jobs:
echo "See the uploaded \`junit-${{ matrix.suite }}\` artifact for details." >> "$GITHUB_STEP_SUMMARY"
exit 1
fi
comment:
name: PR test summary
needs: test
if: >-
always()
&& github.event_name == 'pull_request'
&& github.event.pull_request.head.repo.full_name == github.repository
runs-on: ubuntu-latest
permissions:
pull-requests: write
actions: read
steps:
- name: Download JUnit artifacts
uses: actions/download-artifact@v4
continue-on-error: true
with:
path: junit
pattern: junit-*
- name: Build comment body
run: |
python3 <<'PY'
import glob, os, xml.etree.ElementTree as ET
run_url = f"{os.environ['GITHUB_SERVER_URL']}/{os.environ['GITHUB_REPOSITORY']}/actions/runs/{os.environ['GITHUB_RUN_ID']}"
marker = "<!-- browseros-agent-tests-summary -->"
suites = []
failed_cases = []
total_tests = total_failed = total_skipped = 0
for xml_path in sorted(glob.glob("junit/junit-*/*.xml")):
suite_name = os.path.basename(os.path.dirname(xml_path)).removeprefix("junit-")
try:
root = ET.parse(xml_path).getroot()
except ET.ParseError:
suites.append({"name": suite_name, "passed": 0, "failed": 1, "skipped": 0, "total": 1})
total_tests += 1
total_failed += 1
failed_cases.append((suite_name, "(could not parse junit XML)"))
continue
testsuites = root.findall("testsuite") if root.tag == "testsuites" else [root]
s_tests = s_fail = s_err = s_skip = 0
for ts in testsuites:
s_tests += int(ts.get("tests") or 0)
s_fail += int(ts.get("failures") or 0)
s_err += int(ts.get("errors") or 0)
s_skip += int(ts.get("skipped") or 0)
for tc in ts.iter("testcase"):
if tc.find("failure") is not None or tc.find("error") is not None:
cls = tc.get("classname") or ""
name = tc.get("name") or "(unnamed)"
label = f"{cls} > {name}" if cls else name
failed_cases.append((suite_name, label))
s_failed = s_fail + s_err
s_passed = max(s_tests - s_failed - s_skip, 0)
suites.append({"name": suite_name, "passed": s_passed, "failed": s_failed, "skipped": s_skip, "total": s_tests})
total_tests += s_tests
total_failed += s_failed
total_skipped += s_skip
total_passed = max(total_tests - total_failed - total_skipped, 0)
if total_tests == 0:
header = "## :warning: No test results were produced"
elif total_failed == 0:
header = f"## :white_check_mark: Tests passed — {total_passed}/{total_tests}"
else:
header = f"## :x: Tests failed — {total_failed}/{total_tests} failed"
lines = [marker, header, ""]
if suites:
lines.append("| Suite | Passed | Failed | Skipped |")
lines.append("|-------|--------|--------|---------|")
for s in suites:
icon = ":white_check_mark:" if s["failed"] == 0 and s["total"] > 0 else ":warning:" if s["total"] == 0 else ":x:"
lines.append(f"| {icon} `{s['name']}` | {s['passed']}/{s['total']} | {s['failed']} | {s['skipped']} |")
if failed_cases:
lines += ["", "<details open>", "<summary><b>Failed tests</b></summary>", ""]
for suite_name, label in failed_cases[:50]:
lines.append(f"- **{suite_name}** — `{label}`")
if len(failed_cases) > 50:
lines.append(f"- …and {len(failed_cases) - 50} more")
lines += ["", "</details>"]
lines += ["", f"[View workflow run]({run_url})"]
with open("comment.md", "w") as f:
f.write("\n".join(lines) + "\n")
PY
- name: Upsert sticky PR comment
uses: actions/github-script@v7
with:
script: |
const fs = require('fs');
const body = fs.readFileSync('comment.md', 'utf8');
const marker = '<!-- browseros-agent-tests-summary -->';
const { owner, repo } = context.repo;
const issue_number = context.payload.pull_request.number;
const triggerSha = context.payload.pull_request.head.sha;
const { data: pr } = await github.rest.pulls.get({ owner, repo, pull_number: issue_number });
if (pr.head.sha !== triggerSha) {
core.info(`PR head has moved (${pr.head.sha} vs ${triggerSha}) — skipping stale comment.`);
return;
}
const comments = await github.paginate(github.rest.issues.listComments, {
owner, repo, issue_number, per_page: 100,
});
const existing = comments.find(c => c.body && c.body.includes(marker));
if (existing) {
await github.rest.issues.updateComment({ owner, repo, comment_id: existing.id, body });
} else {
await github.rest.issues.createComment({ owner, repo, issue_number, body });
}

2
.gitignore vendored
View File

@@ -1,4 +1,6 @@
**/.DS_Store
**.auctor/**
.auctor.json
.gcs_entries
**/dmg
**/env

View File

@@ -14,6 +14,7 @@ lerna-debug.log*
# Ignore all .env files except .env.example
**/.env.*
!**/.env.example
!**/.env.sample
!**/.env.production.example
@@ -179,6 +180,8 @@ packages/*/dist
browseros-server
browseros-server.exe
browseros-server-*
tools/alpha/balpha
tools/alpha/browseros-alpha
tools/dev/browseros-dev
log.txt

View File

@@ -218,3 +218,9 @@ This uses the same element resolution as the server's MCP tools — no coordinat
The `<target>` argument can be:
- An **index** from the `targets` output (e.g., `3`)
- A **URL substring** (e.g., `sidepanel`, `newtab`, `chrome-extension://`)
## Release gating — bundled-VM runtime migration (2026-Q2)
Between the Lima server-prod-resources cutover (WS3) and the ContainerRuntime migration (WS6) landing, `resources/bin/third_party/` ships `limactl` instead of `podman`. The current OpenClaw runtime (`apps/server/src/api/services/openclaw/podman-runtime.ts`, `container-runtime.ts`) still invokes `podman`; it will fail to find the binary on builds cut from `dev`.
Do **not** cut a release branch off `dev` during this window. Track WS6 progress before any release cut. See `specs/bundled-vm-runtime-spec.md` + `specs/workstreams.md` for context.

View File

@@ -75,26 +75,20 @@ packages/
### Setup
Requires [process-compose](https://github.com/F1bonacc1/process-compose):
```bash
brew install process-compose
```
```bash
# Copy environment files for each package
cp apps/server/.env.example apps/server/.env.development
cp apps/agent/.env.example apps/agent/.env.development
cp apps/server/.env.production.example apps/server/.env.production
# Install deps, generate agent code, and sync the VM cache
bun run dev:setup
# Start the full dev environment
process-compose up
bun run dev:watch
```
The `process-compose up` command runs the following in order:
1. `bun install` — installs dependencies
2. `bun --cwd apps/agent codegen` — generates agent code
3. `bun --cwd apps/server start` and `bun --cwd apps/agent dev` — starts server and agent in parallel
`dev:watch` exits when the VM cache manifest is missing, but setup stays in `dev:setup`.
### Environment Variables

View File

@@ -74,6 +74,18 @@ const primaryNavItems: NavItem[] = [
{ name: 'Settings', to: '/settings/ai', icon: Settings },
]
function isNavItemActive(item: NavItem, pathname: string): boolean {
if (item.to === '/settings/ai') {
return pathname.startsWith('/settings')
}
if (item.to === '/agents') {
return pathname === '/agents' || pathname.startsWith('/agents/')
}
return pathname === item.to
}
export const SidebarNavigation: FC<SidebarNavigationProps> = ({
expanded = true,
}) => {
@@ -90,10 +102,7 @@ export const SidebarNavigation: FC<SidebarNavigationProps> = ({
<nav className="space-y-1">
{filteredItems.map((item) => {
const Icon = item.icon
const isActive =
item.to === '/settings/ai'
? location.pathname.startsWith('/settings')
: location.pathname === item.to
const isActive = isNavItemActive(item, location.pathname)
const navItem = (
<NavLink

View File

@@ -113,7 +113,22 @@ export const App: FC = () => {
<Route path="connect-apps" element={<ConnectMCP />} />
<Route path="scheduled" element={<ScheduledTasksPage />} />
{alphaEnabled ? (
<Route path="agents" element={<AgentsPage />} />
<>
<Route path="agents" element={<AgentsPage />} />
<Route element={<AgentCommandLayout />}>
<Route
path="agents/:agentId"
element={
<AgentCommandConversation
variant="page"
backPath="/agents"
agentPathPrefix="/agents"
createAgentPath="/agents"
/>
}
/>
</Route>
</>
) : null}
{alphaEnabled ? (
<Route path="admin" element={<AdminDashboardPage />} />

View File

@@ -1,4 +1,4 @@
import { Bot } from 'lucide-react'
import { Bot, Loader2, Wrench } from 'lucide-react'
import type { FC } from 'react'
import type { AgentCardData } from '@/lib/agent-conversations/types'
import { cn } from '@/lib/utils'
@@ -32,6 +32,11 @@ function getStatusTone(status: AgentCardData['status']): string {
return 'bg-emerald-500'
}
function formatCost(usd: number): string {
if (usd < 0.005) return `$${usd.toFixed(4)}`
return `$${usd.toFixed(2)}`
}
export const AgentCardExpanded: FC<AgentCardProps> = ({
agent,
onClick,
@@ -81,9 +86,26 @@ export const AgentCardExpanded: FC<AgentCardProps> = ({
</p>
</div>
<div className="mt-4 flex items-center justify-between gap-3 text-muted-foreground text-xs">
<span>{formatTimestamp(agent.lastMessageTimestamp)}</span>
<span>Open conversation</span>
<div className="mt-4 space-y-1.5 text-muted-foreground text-xs">
<div className="flex items-center justify-between gap-3">
<span>{formatTimestamp(agent.lastMessageTimestamp)}</span>
{agent.costUsd ? (
<span className="tabular-nums opacity-70">
{formatCost(agent.costUsd)}
</span>
) : null}
</div>
{agent.status === 'working' && agent.currentTool ? (
<div className="flex items-center gap-1.5 text-[var(--accent-orange)]/70">
<Loader2 className="size-3 shrink-0 animate-spin" />
<span className="truncate">{agent.currentTool}</span>
</div>
) : agent.activitySummary ? (
<div className="flex items-center gap-1.5 text-muted-foreground/60">
<Wrench className="size-3 shrink-0" />
<span className="truncate">{agent.activitySummary}</span>
</div>
) : null}
</div>
</button>
)

View File

@@ -1,194 +1,429 @@
import { Bot, Home, RotateCcw } from 'lucide-react'
import { type FC, useEffect, useRef } from 'react'
import { ArrowLeft, Bot, Home } from 'lucide-react'
import { type FC, useEffect, useMemo, useRef, useState } from 'react'
import { Navigate, useNavigate, useParams, useSearchParams } from 'react-router'
import { Button } from '@/components/ui/button'
import type { AgentEntry } from '@/entrypoints/app/agents/useOpenClaw'
import {
type AgentEntry,
getModelDisplayName,
} from '@/entrypoints/app/agents/useOpenClaw'
import { cn } from '@/lib/utils'
import { useAgentCommandData } from './agent-command-layout'
import { ClawChat } from './ClawChat'
import { ConversationInput } from './ConversationInput'
import { ConversationMessage } from './ConversationMessage'
import {
buildChatHistoryFromClawMessages,
flattenHistoryPages,
} from './claw-chat-types'
import { useAgentConversation } from './useAgentConversation'
import { useClawChatHistory } from './useClawChatHistory'
import { useOutboundQueue } from './useOutboundQueue'
function StatusBadge({ status }: { status: string }) {
return (
<div className="inline-flex items-center gap-2 rounded-full border border-border/60 bg-card px-3 py-1 text-[11px] text-muted-foreground uppercase tracking-[0.18em]">
<span
className={cn(
'size-1.5 rounded-full',
status === 'Working on your request'
? 'bg-amber-500'
: status === 'Ready'
? 'bg-emerald-500'
: status === 'Offline'
? 'bg-muted-foreground/50'
: 'bg-[var(--accent-orange)]',
)}
/>
<span>{status}</span>
</div>
)
}
function AgentIdentity({
name,
meta,
className,
}: {
name: string
meta: string
className?: string
}) {
return (
<div className={cn('min-w-0', className)}>
<div className="truncate font-semibold text-[15px] leading-5">{name}</div>
<div className="truncate text-muted-foreground text-xs leading-5">
{meta}
</div>
</div>
)
}
function ConversationHeader({
agentName,
agentMeta,
status,
backLabel,
backTarget,
onGoHome,
onReset,
}: {
agentName: string
agentMeta: string
status: string
backLabel: string
backTarget: 'home' | 'page'
onGoHome: () => void
onReset: () => void
}) {
const BackIcon = backTarget === 'home' ? Home : ArrowLeft
return (
<div className="overflow-hidden rounded-[1.5rem] border border-border/60 bg-card/95 shadow-sm backdrop-blur">
<div className="flex items-center justify-between gap-3 px-5 py-4">
<div className="flex min-w-0 items-center gap-3">
<Button
variant="ghost"
size="icon"
onClick={onGoHome}
className="rounded-xl"
title="Back to home"
>
<Home className="size-4" />
</Button>
<div className="flex size-11 shrink-0 items-center justify-center rounded-2xl bg-muted text-muted-foreground">
<Bot className="size-5" />
</div>
<div className="min-w-0">
<div className="truncate font-semibold text-sm">{agentName}</div>
<div className="truncate text-muted-foreground text-sm">
{status}
</div>
</div>
</div>
<div className="flex h-14 items-center justify-between gap-4 border-border/50 border-b px-5">
<div className="flex min-w-0 items-center gap-3">
<Button
variant="ghost"
size="sm"
onClick={onReset}
className="rounded-xl text-muted-foreground"
size="icon"
onClick={onGoHome}
className="size-8 rounded-xl lg:hidden"
title={backLabel}
>
<RotateCcw className="mr-2 size-4" />
New conversation
<BackIcon className="size-4" />
</Button>
</div>
</div>
)
}
function EmptyConversationState({ agentName }: { agentName: string }) {
return (
<div className="flex min-h-full items-center justify-center py-10">
<div className="max-w-md rounded-[1.5rem] border border-border/60 bg-card/90 px-8 py-10 text-center shadow-sm backdrop-blur">
<div className="mx-auto flex size-14 items-center justify-center rounded-2xl bg-muted text-muted-foreground">
<Bot className="size-6" />
<div className="flex size-8 shrink-0 items-center justify-center rounded-xl bg-muted text-muted-foreground">
<Bot className="size-4" />
</div>
<AgentIdentity name={agentName} meta={agentMeta} />
</div>
<StatusBadge status={status} />
</div>
)
}
function AgentRailHeader({ onGoHome }: { onGoHome: () => void }) {
return (
<div className="hidden h-14 items-center border-border/50 border-r border-b bg-background/70 px-4 lg:flex">
<div className="flex min-w-0 items-center gap-3">
<Button
variant="ghost"
size="icon"
onClick={onGoHome}
className="size-8 rounded-xl"
title="Back to home"
>
<ArrowLeft className="size-4" />
</Button>
<div className="truncate font-semibold text-[15px] leading-5">
Agents
</div>
<h2 className="mt-4 font-semibold text-lg">{agentName}</h2>
<p className="mt-2 text-muted-foreground text-sm">
Send a message to start a focused conversation with this agent.
</p>
</div>
</div>
)
}
function getConversationStatusCopy(
status: string | undefined,
streaming: boolean,
): string {
if (streaming) return 'Working on your request'
if (status === 'running') return 'Ready for the next task'
if (status === 'starting') return 'Connecting to OpenClaw'
if (status === 'error') return 'OpenClaw needs attention'
if (status === 'stopped') return 'OpenClaw is offline'
return 'Open agent setup to continue'
function AgentRailList({
activeAgentId,
agents,
onSelectAgent,
}: {
activeAgentId: string
agents: AgentEntry[]
onSelectAgent: (entry: AgentEntry) => void
}) {
return (
<aside className="hidden min-h-0 flex-col border-border/50 border-r bg-background/70 lg:flex">
<div className="styled-scrollbar min-h-0 flex-1 space-y-2 overflow-y-auto px-3 py-3">
{agents.map((entry) => {
const active = entry.agentId === activeAgentId
const modelName = getModelDisplayName(entry.model) ?? 'OpenClaw agent'
return (
<button
key={entry.agentId}
type="button"
onClick={() => onSelectAgent(entry)}
className={cn(
'w-full rounded-2xl border px-3 py-3 text-left transition-all',
active
? 'border-[var(--accent-orange)]/30 bg-[var(--accent-orange)]/8 shadow-sm'
: 'border-transparent bg-transparent hover:border-border/60 hover:bg-card',
)}
>
<div className="flex items-center gap-3">
<div
className={cn(
'flex size-9 items-center justify-center rounded-xl',
active
? 'bg-[var(--accent-orange)]/12 text-[var(--accent-orange)]'
: 'bg-muted text-muted-foreground',
)}
>
<Bot className="size-4" />
</div>
<AgentIdentity name={entry.name} meta={modelName} />
</div>
</button>
)
})}
</div>
</aside>
)
}
export const AgentCommandConversation: FC = () => {
const { agentId } = useParams<{ agentId: string }>()
const [searchParams, setSearchParams] = useSearchParams()
function getConversationStatusCopy(status: string | undefined): string {
if (status === 'running') return 'Ready'
if (status === 'starting') return 'Connecting'
if (status === 'error') return 'Attention'
if (status === 'stopped') return 'Offline'
return 'Setup'
}
function AgentConversationController({
agentId,
initialMessage,
onInitialMessageConsumed,
status,
agents,
agentPathPrefix,
createAgentPath,
}: {
agentId: string
initialMessage: string | null
onInitialMessageConsumed: () => void
status: ReturnType<typeof useAgentCommandData>['status']
agents: AgentEntry[]
agentPathPrefix: string
createAgentPath: string
}) {
const navigate = useNavigate()
const scrollRef = useRef<HTMLDivElement>(null)
const initialQuerySent = useRef(false)
const { status, agents } = useAgentCommandData()
const shouldRedirectHome = !agentId
const resolvedAgentId = agentId ?? ''
const agent = agents.find((entry) => entry.agentId === resolvedAgentId)
const agentName = agent?.name || resolvedAgentId || 'Agent'
const { turns, streaming, loading, send, resetConversation } =
useAgentConversation(resolvedAgentId, agentName)
const lastTurn = turns[turns.length - 1]
const lastTurnPartCount = lastTurn?.parts.length ?? 0
const initialMessageSentRef = useRef<string | null>(null)
const onInitialMessageConsumedRef = useRef(onInitialMessageConsumed)
const [streamSessionKey, setStreamSessionKey] = useState<string | null>(null)
const agent = agents.find((entry) => entry.agentId === agentId)
const agentName = agent?.name || agentId || 'Agent'
// Single source of truth: the history endpoint resolves the session itself
// when sessionKey is null. Once a chat creates a new session, streamSessionKey
// overrides it and the history queryKey rotates to refetch for that session.
const historyQuery = useClawChatHistory({
agentId,
sessionKey: streamSessionKey,
})
const historyMessages = useMemo(
() => flattenHistoryPages(historyQuery.data?.pages ?? []),
[historyQuery.data?.pages],
)
const chatHistory = useMemo(
() => buildChatHistoryFromClawMessages(historyMessages),
[historyMessages],
)
const resolvedSessionKey =
streamSessionKey ?? historyQuery.data?.pages?.[0]?.sessionKey ?? null
const { turns, streaming } = useAgentConversation(agentId, {
sessionKey: resolvedSessionKey,
history: chatHistory,
onSessionKeyChange: (sessionKey) => {
setStreamSessionKey(sessionKey)
},
})
const outboundQueue = useOutboundQueue({
agentId,
sessionKey: resolvedSessionKey,
})
onInitialMessageConsumedRef.current = onInitialMessageConsumed
// Refetch history whenever a server-dispatched queue item completes.
// The server worker streams the queued turn into OpenClaw directly, so
// the client never observes the live tokens — we only see the new
// assistant turn once the JSONL is updated. Watching the queue for
// any 'sending' item dropping out is the cleanest "turn finalized"
// signal we have without exposing per-turn SSE.
const previousSendingIdsRef = useRef<Set<string>>(new Set())
useEffect(() => {
if (shouldRedirectHome) return
const query = searchParams.get('q')
if (query && !initialQuerySent.current && !loading) {
initialQuerySent.current = true
setSearchParams({}, { replace: true })
void send(query)
const currentSending = new Set(
outboundQueue.queue
.filter((item) => item.status === 'sending')
.map((item) => item.id),
)
const dropped = [...previousSendingIdsRef.current].filter(
(id) => !currentSending.has(id),
)
previousSendingIdsRef.current = currentSending
if (dropped.length > 0) {
void historyQuery.refetch()
}
}, [loading, searchParams, send, setSearchParams, shouldRedirectHome])
}, [outboundQueue.queue, historyQuery])
const disabled = status?.status !== 'running'
// Two-part gate: cover both "still fetching" AND "just got enabled but
// hasn't started fetching yet". When `enabled` flips true (baseUrl
// resolves), there's a render frame where React Query reports
// isLoading=false but hasn't run the queryFn yet — `isFetched` is still
// false. Without this we render EmptyState during that one frame.
const isInitialLoading =
historyQuery.isLoading || (!historyQuery.isFetched && !historyQuery.isError)
const historyReady = historyQuery.isFetched || historyQuery.isError
const initialMessageKey = initialMessage
? `${agentId}:${initialMessage}`
: null
const error = historyQuery.error ?? null
const enqueueRef = useRef(outboundQueue.enqueue)
enqueueRef.current = outboundQueue.enqueue
useEffect(() => {
const query = initialMessage?.trim()
if (!initialMessageKey) {
initialMessageSentRef.current = null
return
}
// The initial-message handoff (home composer → conversation page via
// ?q=) goes through the outbound queue too, so it inherits the same
// single-flight serialization. We no longer need to gate on
// `streaming` — the queue worker drains as soon as the agent is
// free.
if (
shouldRedirectHome ||
(turns.length === 0 && lastTurnPartCount === 0 && !streaming)
!query ||
initialMessageSentRef.current === initialMessageKey ||
disabled ||
!historyReady
) {
return
}
scrollRef.current?.scrollTo({
top: scrollRef.current.scrollHeight,
behavior: 'smooth',
})
}, [lastTurnPartCount, shouldRedirectHome, streaming, turns.length])
if (shouldRedirectHome) {
return <Navigate to="/home" replace />
}
initialMessageSentRef.current = initialMessageKey
onInitialMessageConsumedRef.current()
enqueueRef.current({ text: query })
}, [disabled, historyReady, initialMessage, initialMessageKey])
const handleSelectAgent = (entry: AgentEntry) => {
navigate(`/home/agents/${entry.agentId}`)
navigate(`${agentPathPrefix}/${entry.agentId}`)
}
const statusCopy = getConversationStatusCopy(status?.status, streaming)
return (
<div className="absolute inset-0 overflow-hidden">
<div className="fade-in slide-in-from-bottom-5 mx-auto flex h-full w-full max-w-3xl animate-in flex-col gap-3 px-4 pt-4 pb-2 duration-300">
<ConversationHeader
agentName={agentName}
status={statusCopy}
onGoHome={() => navigate('/home')}
onReset={resetConversation}
/>
<div className="flex min-h-0 flex-col overflow-hidden">
<ClawChat
agentName={agentName}
historyMessages={historyMessages}
turns={turns}
streaming={streaming}
isInitialLoading={isInitialLoading}
error={error}
hasNextPage={Boolean(historyQuery.hasNextPage)}
isFetchingNextPage={historyQuery.isFetchingNextPage}
onFetchNextPage={() => {
void historyQuery.fetchNextPage()
}}
onRetry={() => {
void historyQuery.refetch()
}}
/>
<main
ref={scrollRef}
className={cn(
'styled-scrollbar min-h-0 flex-1 overflow-y-auto overflow-x-hidden rounded-[1.5rem] border border-border/50 bg-card/85 px-5 py-5 shadow-sm',
'[&_[data-streamdown="code-block"]]:!max-w-full [&_[data-streamdown="table-wrapper"]]:!max-w-full [&_[data-streamdown="code-block"]]:overflow-x-auto [&_[data-streamdown="table-wrapper"]]:overflow-x-auto',
)}
>
{loading ? (
<div className="flex h-full items-center justify-center text-muted-foreground text-sm">
Loading conversation...
</div>
) : turns.length === 0 ? (
<EmptyConversationState agentName={agentName} />
) : (
<div className="w-full space-y-4">
{turns.map((turn, index) => (
<ConversationMessage
key={turn.id}
turn={turn}
streaming={streaming && index === turns.length - 1}
/>
))}
</div>
)}
</main>
<div className="w-full flex-shrink-0">
<div className="border-border/50 border-t bg-background/88 px-4 py-3 backdrop-blur-md">
<div className="mx-auto max-w-3xl">
<ConversationInput
variant="conversation"
agents={agents}
selectedAgentId={resolvedAgentId}
selectedAgentId={agentId}
onSelectAgent={handleSelectAgent}
onSend={(text) => {
void send(text)
onSend={(input) => {
outboundQueue.enqueue({
text: input.text,
attachments: input.attachments.map((a) => a.payload),
attachmentPreviews: input.attachments.map((a) => ({
id: a.id,
kind: a.kind,
mediaType: a.mediaType,
name: a.name,
dataUrl: a.dataUrl,
})),
history: chatHistory,
})
}}
onCreateAgent={() => navigate('/agents')}
onCreateAgent={() => navigate(createAgentPath)}
streaming={streaming}
disabled={status?.status !== 'running'}
disabled={disabled}
status={status?.status}
placeholder={`Message ${agentName}...`}
outboundQueue={outboundQueue.queue}
onCancelQueued={outboundQueue.cancel}
onRetryQueued={outboundQueue.retry}
/>
</div>
</div>
</div>
)
}
interface AgentCommandConversationProps {
variant?: 'command' | 'page'
backPath?: string
agentPathPrefix?: string
createAgentPath?: string
}
export const AgentCommandConversation: FC<AgentCommandConversationProps> = ({
variant = 'command',
backPath = '/home',
agentPathPrefix = '/home/agents',
createAgentPath = '/agents',
}) => {
const { agentId } = useParams<{ agentId: string }>()
const [searchParams, setSearchParams] = useSearchParams()
const navigate = useNavigate()
const { status, agents } = useAgentCommandData()
const shouldRedirectHome = !agentId
const resolvedAgentId = agentId ?? ''
const agent = agents.find((entry) => entry.agentId === resolvedAgentId)
const agentName = agent?.name || resolvedAgentId || 'Agent'
const agentMeta = getModelDisplayName(agent?.model) ?? 'OpenClaw agent'
const initialMessage = searchParams.get('q')
const isPageVariant = variant === 'page'
const backLabel = isPageVariant ? 'Back to agents' : 'Back to home'
if (shouldRedirectHome) {
return <Navigate to="/home" replace />
}
const handleSelectAgent = (entry: AgentEntry) => {
navigate(`${agentPathPrefix}/${entry.agentId}`)
}
const statusCopy = getConversationStatusCopy(status?.status)
return (
<div className="absolute inset-0 overflow-hidden bg-background md:pl-[theme(spacing.14)]">
<div className="mx-auto grid h-full w-full max-w-[1480px] lg:grid-cols-[288px_minmax(0,1fr)] lg:grid-rows-[3.5rem_minmax(0,1fr)]">
<AgentRailHeader onGoHome={() => navigate(backPath)} />
<ConversationHeader
agentName={agentName}
agentMeta={agentMeta}
status={statusCopy}
backLabel={backLabel}
backTarget={isPageVariant ? 'page' : 'home'}
onGoHome={() => navigate(backPath)}
/>
<AgentRailList
activeAgentId={resolvedAgentId}
agents={agents}
onSelectAgent={handleSelectAgent}
/>
<AgentConversationController
key={resolvedAgentId}
agentId={resolvedAgentId}
agents={agents}
status={status}
initialMessage={initialMessage}
onInitialMessageConsumed={() =>
setSearchParams({}, { replace: true })
}
agentPathPrefix={agentPathPrefix}
createAgentPath={createAgentPath}
/>
</div>
</div>
)
}

View File

@@ -1,20 +1,19 @@
import { ArrowRight } from 'lucide-react'
import { ArrowRight, Bot, Plus, Settings2 } from 'lucide-react'
import { type FC, useEffect, useState } from 'react'
import { useNavigate } from 'react-router'
import { Button } from '@/components/ui/button'
import { Card, CardContent } from '@/components/ui/card'
import { Separator } from '@/components/ui/separator'
import type { AgentEntry } from '@/entrypoints/app/agents/useOpenClaw'
import { ImportDataHint } from '@/entrypoints/newtab/index/ImportDataHint'
import { NewTabBranding } from '@/entrypoints/newtab/index/NewTabBranding'
import { NewTabTip } from '@/entrypoints/newtab/index/NewTabTip'
import { ScheduleResults } from '@/entrypoints/newtab/index/ScheduleResults'
import { SignInHint } from '@/entrypoints/newtab/index/SignInHint'
import { TopSites } from '@/entrypoints/newtab/index/TopSites'
import { useActiveHint } from '@/entrypoints/newtab/index/useActiveHint'
import type { AgentCardData } from '@/lib/agent-conversations/types'
import { AgentCardDock } from './AgentCardDock'
import { useAgentCommandData } from './agent-command-layout'
import { ConversationInput } from './ConversationInput'
import { useAgentCardData } from './useAgentCardData'
import { buildAgentCardData } from './useAgentCardData'
import { useAgentDashboard } from './useAgentDashboard'
function AgentCommandSetupState({
onOpenAgents,
@@ -22,13 +21,19 @@ function AgentCommandSetupState({
onOpenAgents: () => void
}) {
return (
<Card className="border-border/60 bg-card/85 shadow-sm">
<CardContent className="flex flex-col items-center gap-4 p-6 text-center">
<p className="max-w-xl text-muted-foreground text-sm">
Set up OpenClaw agents to turn your new tab into an agent command
center.
</p>
<Button onClick={onOpenAgents} className="gap-2">
<Card className="border-border/60 bg-card/90 shadow-sm">
<CardContent className="flex flex-col items-center gap-4 p-8 text-center">
<div className="flex size-12 items-center justify-center rounded-2xl bg-muted text-muted-foreground">
<Bot className="size-5" />
</div>
<div className="space-y-2">
<h2 className="font-semibold text-lg">Set up your first agent</h2>
<p className="max-w-md text-muted-foreground text-sm leading-6">
Connect OpenClaw and create an agent before using the new tab as
your workspace.
</p>
</div>
<Button onClick={onOpenAgents} className="gap-2 rounded-xl">
Open Agent Setup
<ArrowRight className="size-4" />
</Button>
@@ -39,13 +44,19 @@ function AgentCommandSetupState({
function EmptyAgentsState({ onOpenAgents }: { onOpenAgents: () => void }) {
return (
<Card className="border-border/60 bg-card/85 shadow-sm">
<CardContent className="flex flex-col items-center gap-4 p-6 text-center">
<p className="max-w-xl text-muted-foreground text-sm">
OpenClaw is running, but you do not have any agents yet.
</p>
<Button variant="outline" onClick={onOpenAgents}>
Create your first agent
<Card className="border-border/60 bg-card/90 shadow-sm">
<CardContent className="flex flex-col items-center gap-4 p-8 text-center">
<div className="flex size-12 items-center justify-center rounded-2xl bg-muted text-muted-foreground">
<Plus className="size-5" />
</div>
<div className="space-y-2">
<h2 className="font-semibold text-lg">No agents yet</h2>
<p className="max-w-md text-muted-foreground text-sm leading-6">
Create an agent to start using BrowserOS as an agent-first new tab.
</p>
</div>
<Button variant="outline" onClick={onOpenAgents} className="rounded-xl">
Create agent
</Button>
</CardContent>
</Card>
@@ -58,13 +69,19 @@ function OpenClawUnavailableState({
onOpenAgents: () => void
}) {
return (
<Card className="border-border/60 bg-card/85 shadow-sm">
<CardContent className="flex flex-col items-center gap-4 p-6 text-center">
<p className="max-w-xl text-muted-foreground text-sm">
OpenClaw is unavailable right now. Open the Agents page to restart the
gateway or review setup.
</p>
<Button onClick={onOpenAgents} className="gap-2">
<Card className="border-border/60 bg-card/90 shadow-sm">
<CardContent className="flex flex-col items-center gap-4 p-8 text-center">
<div className="flex size-12 items-center justify-center rounded-2xl bg-muted text-muted-foreground">
<Settings2 className="size-5" />
</div>
<div className="space-y-2">
<h2 className="font-semibold text-lg">OpenClaw is unavailable</h2>
<p className="max-w-md text-muted-foreground text-sm leading-6">
Review your agent setup to restart the gateway or reconnect the
local service.
</p>
</div>
<Button onClick={onOpenAgents} className="gap-2 rounded-xl">
Open Agent Setup
<ArrowRight className="size-4" />
</Button>
@@ -73,17 +90,54 @@ function OpenClawUnavailableState({
)
}
function RecentThreads({
activeAgentId,
agents,
onOpenAgents,
onSelectAgent,
}: {
activeAgentId?: string | null
agents: AgentCardData[]
onOpenAgents: () => void
onSelectAgent: (agentId: string) => void
}) {
if (agents.length === 0) return null
return (
<section className="space-y-4">
<div className="flex items-center justify-between gap-4">
<div>
<h2 className="font-semibold text-base">Recent agents</h2>
<p className="text-muted-foreground text-sm">
Continue from where you left off.
</p>
</div>
<Button
variant="outline"
onClick={onOpenAgents}
className="rounded-xl"
size="sm"
>
Manage agents
</Button>
</div>
<AgentCardDock
agents={agents}
activeAgentId={activeAgentId ?? undefined}
onSelectAgent={onSelectAgent}
onCreateAgent={onOpenAgents}
/>
</section>
)
}
export const AgentCommandHome: FC = () => {
const navigate = useNavigate()
const activeHint = useActiveHint()
const { status, agents } = useAgentCommandData()
const [mounted, setMounted] = useState(false)
const [selectedAgentId, setSelectedAgentId] = useState<string | null>(null)
const cardData = useAgentCardData(agents, status?.status)
useEffect(() => {
setMounted(true)
}, [])
const { data: dashboard } = useAgentDashboard(status?.status === 'running')
const cardData = buildAgentCardData(agents, status?.status, dashboard?.agents)
useEffect(() => {
if (agents.length === 0) {
@@ -101,9 +155,16 @@ export const AgentCommandHome: FC = () => {
}
}, [agents, selectedAgentId])
const handleSend = (text: string) => {
const handleSend = (input: { text: string }) => {
if (!selectedAgentId) return
navigate(`/home/agents/${selectedAgentId}?q=${encodeURIComponent(text)}`)
// Home composer navigates to the conversation page with the prompt in
// the query string. Attachments are dropped at this boundary in v1 —
// the conversation page (where staging UX is most useful anyway) is
// where users can attach. A future iteration can stash staged files
// in chrome.storage.session and replay them on first mount there.
navigate(
`/home/agents/${selectedAgentId}?q=${encodeURIComponent(input.text)}`,
)
}
const handleSelectAgent = (agent: AgentEntry) => {
@@ -117,62 +178,65 @@ export const AgentCommandHome: FC = () => {
openClawStatus !== 'running' &&
openClawStatus !== 'uninitialized' &&
cardData.length === 0
const selectedCard =
cardData.find((agent) => agent.agentId === selectedAgentId) ?? cardData[0]
return (
<div className="pt-[max(25vh,16px)]">
<div className="relative w-full space-y-8 md:w-3xl">
<NewTabBranding />
<ConversationInput
variant="home"
agents={agents}
selectedAgentId={selectedAgentId}
onSelectAgent={handleSelectAgent}
onSend={handleSend}
onCreateAgent={() => navigate('/agents')}
streaming={false}
disabled={status?.status !== 'running'}
status={status?.status}
placeholder={
status?.status === 'running'
? undefined
: 'OpenClaw is not running...'
}
/>
{mounted ? <NewTabTip /> : null}
<div className="min-h-full px-4 py-6">
<div className="mx-auto flex w-full max-w-5xl flex-col gap-8">
{isSetup ? (
shouldShowUnavailableState ? (
<OpenClawUnavailableState
onOpenAgents={() => navigate('/agents')}
/>
) : cardData.length > 0 ? (
<section className="space-y-3">
<div className="flex items-center justify-between">
<div>
<h2 className="font-semibold text-base">Agents</h2>
<p className="text-muted-foreground text-sm">
Pick up where your agents left off.
<>
<div className="flex flex-col items-center gap-5 pt-[max(10vh,24px)] text-center">
<div className="space-y-3">
<h1 className="font-semibold text-[clamp(2rem,4vw,3.25rem)] leading-tight tracking-tight">
What should your agent work on next?
</h1>
<p className="mx-auto max-w-2xl text-muted-foreground text-sm leading-6">
Start with a task, continue a thread, or switch to another
agent without leaving the new tab.
</p>
</div>
<div className="w-full max-w-3xl">
<ConversationInput
variant="home"
agents={agents}
selectedAgentId={selectedAgentId}
onSelectAgent={handleSelectAgent}
onSend={handleSend}
onCreateAgent={() => navigate('/agents')}
streaming={false}
disabled={status?.status !== 'running'}
status={status?.status}
placeholder={
status?.status === 'running'
? `Ask ${selectedCard?.name ?? 'your agent'} to handle a task...`
: 'OpenClaw is not running...'
}
/>
</div>
</div>
<AgentCardDock
<Separator />
<RecentThreads
activeAgentId={selectedAgentId}
agents={cardData}
activeAgentId={selectedAgentId ?? undefined}
onOpenAgents={() => navigate('/agents')}
onSelectAgent={(agentId) => navigate(`/home/agents/${agentId}`)}
onCreateAgent={() => navigate('/agents')}
/>
</section>
</>
) : (
<EmptyAgentsState onOpenAgents={() => navigate('/agents')} />
)
) : (
<AgentCommandSetupState onOpenAgents={() => navigate('/agents')} />
)}
{mounted ? <TopSites /> : null}
{mounted ? <ScheduleResults /> : null}
</div>
{activeHint === 'signin' ? <SignInHint /> : null}

View File

@@ -0,0 +1,172 @@
import { Bot, Loader2, RefreshCw } from 'lucide-react'
import { type FC, useEffect, useRef } from 'react'
import {
Conversation,
ConversationContent,
ConversationScrollButton,
} from '@/components/ai-elements/conversation'
import type { AgentConversationTurn } from '@/lib/agent-conversations/types'
import { cn } from '@/lib/utils'
import { ClawChatMessage } from './ClawChatMessage'
import { ConversationMessage } from './ConversationMessage'
import type { ClawChatMessage as ClawChatMessageModel } from './claw-chat-types'
interface ClawChatProps {
agentName: string
historyMessages: ClawChatMessageModel[]
turns: AgentConversationTurn[]
streaming: boolean
isInitialLoading: boolean
error: Error | null
hasNextPage: boolean
isFetchingNextPage: boolean
onFetchNextPage: () => void
onRetry: () => void
className?: string
}
function EmptyConversationState({ agentName }: { agentName: string }) {
return (
<div className="flex h-full items-center justify-center px-6 py-12">
<div className="max-w-md text-center">
<div className="mx-auto flex size-14 items-center justify-center rounded-3xl bg-muted text-muted-foreground">
<Bot className="size-6" />
</div>
<h2 className="mt-5 font-semibold text-xl">{agentName}</h2>
<p className="mt-2 text-muted-foreground text-sm leading-6">
Ask {agentName} to start a task.
</p>
</div>
</div>
)
}
function LoadingConversationState() {
return (
<div className="flex h-full items-center justify-center gap-2 text-muted-foreground text-sm">
<Loader2 className="size-4 animate-spin" />
Loading conversation...
</div>
)
}
function ConversationErrorState({
message,
onRetry,
}: {
message: string
onRetry: () => void
}) {
return (
<div className="flex h-full items-center justify-center px-6 py-12">
<div className="max-w-md rounded-2xl border border-border/60 bg-card px-5 py-4 text-center shadow-sm">
<p className="text-sm">{message}</p>
<button
type="button"
onClick={onRetry}
className="mt-3 inline-flex items-center gap-2 rounded-lg border border-border/60 px-3 py-1.5 font-medium text-muted-foreground text-xs transition-colors hover:bg-accent hover:text-foreground"
>
<RefreshCw className="size-3.5" />
Retry
</button>
</div>
</div>
)
}
export const ClawChat: FC<ClawChatProps> = ({
agentName,
historyMessages,
turns,
streaming,
isInitialLoading,
error,
hasNextPage,
isFetchingNextPage,
onFetchNextPage,
onRetry,
className,
}) => {
const topSentinelRef = useRef<HTMLDivElement>(null)
const onFetchNextPageRef = useRef(onFetchNextPage)
onFetchNextPageRef.current = onFetchNextPage
const hasMessages = historyMessages.length > 0 || turns.length > 0
useEffect(() => {
const sentinel = topSentinelRef.current
if (!sentinel) return
const observer = new IntersectionObserver(
(entries) => {
const [entry] = entries
if (!entry?.isIntersecting || !hasNextPage || isFetchingNextPage) {
return
}
onFetchNextPageRef.current()
},
{
root: null,
rootMargin: '160px 0px 0px 0px',
threshold: 0,
},
)
observer.observe(sentinel)
return () => observer.disconnect()
}, [hasNextPage, isFetchingNextPage])
return (
<div
className={cn('flex min-h-0 flex-1 flex-col overflow-hidden', className)}
>
<Conversation
className={cn(
'bg-background',
'[&_[data-streamdown="code-block"]]:!w-full [&_[data-streamdown="code-block"]]:!max-w-full [&_[data-streamdown="table-wrapper"]]:!w-full [&_[data-streamdown="table-wrapper"]]:!max-w-full [&_[data-streamdown="code-block"]]:overflow-x-auto [&_[data-streamdown="table-wrapper"]]:overflow-x-auto',
)}
>
<ConversationContent className="min-h-full px-5 py-5">
{isInitialLoading ? (
<LoadingConversationState />
) : error && !hasMessages ? (
<ConversationErrorState message={error.message} onRetry={onRetry} />
) : !hasMessages ? (
<EmptyConversationState agentName={agentName} />
) : (
<div className="mx-auto flex w-full max-w-3xl flex-col gap-3">
<div ref={topSentinelRef} aria-hidden="true" className="h-px" />
{isFetchingNextPage ? (
<div className="flex justify-center py-2 text-muted-foreground text-xs">
<Loader2 className="mr-2 size-3.5 animate-spin" />
Loading older messages...
</div>
) : null}
{!hasNextPage && historyMessages.length > 0 ? (
<div className="py-1 text-center text-muted-foreground text-xs">
Start of conversation
</div>
) : null}
{historyMessages.map((message) => (
<ClawChatMessage key={message.id} message={message} />
))}
{turns.map((turn, index) => (
<ConversationMessage
key={turn.id}
turn={turn}
streaming={streaming && index === turns.length - 1}
/>
))}
{error ? (
<div className="rounded-xl border border-border/60 bg-card px-4 py-3 text-muted-foreground text-sm">
{error.message}
</div>
) : null}
</div>
)}
</ConversationContent>
<ConversationScrollButton />
</Conversation>
</div>
)
}

View File

@@ -0,0 +1,248 @@
import { CheckCircle2, Copy, Loader2, Wrench, XCircle } from 'lucide-react'
import { type FC, useCallback, useMemo } from 'react'
import {
Message,
MessageAction,
MessageActions,
MessageAttachment,
MessageAttachments,
MessageContent,
MessageResponse,
MessageToolbar,
} from '@/components/ai-elements/message'
import {
Reasoning,
ReasoningContent,
ReasoningTrigger,
} from '@/components/ai-elements/reasoning'
import {
Task,
TaskContent,
TaskItem,
TaskTrigger,
} from '@/components/ai-elements/task'
import { cn } from '@/lib/utils'
import type {
ClawChatMessagePart,
ClawChatMessage as ClawChatMessageType,
} from './claw-chat-types'
function formatCost(usd: number): string {
if (usd < 0.005) return `$${usd.toFixed(4)}`
return `$${usd.toFixed(2)}`
}
type ToolCallPart = Extract<ClawChatMessagePart, { type: 'tool-call' }>
type AttachmentPart = Extract<ClawChatMessagePart, { type: 'attachment' }>
interface RenderEntry {
kind: 'text' | 'reasoning' | 'meta' | 'task' | 'attachments'
partIndex: number
part?: ClawChatMessagePart
tools?: ToolCallPart[]
attachments?: AttachmentPart[]
}
/**
* Build a render plan that groups all tool-call parts into a single Task
* collapsible and all attachment parts into a single attachment strip at
* their respective first-appearance positions. Other parts render in place.
*/
function buildRenderEntries(parts: ClawChatMessagePart[]): RenderEntry[] {
const entries: RenderEntry[] = []
const tools: ToolCallPart[] = []
const attachments: AttachmentPart[] = []
let taskInserted = false
let attachmentsInserted = false
parts.forEach((part, partIndex) => {
if (part.type === 'tool-call') {
tools.push(part)
if (!taskInserted) {
entries.push({ kind: 'task', partIndex, tools })
taskInserted = true
}
} else if (part.type === 'attachment') {
attachments.push(part)
if (!attachmentsInserted) {
entries.push({ kind: 'attachments', partIndex, attachments })
attachmentsInserted = true
}
} else if (part.type === 'text') {
entries.push({ kind: 'text', partIndex, part })
} else if (part.type === 'reasoning') {
entries.push({ kind: 'reasoning', partIndex, part })
} else if (part.type === 'meta') {
entries.push({ kind: 'meta', partIndex, part })
}
})
return entries
}
function ToolStatusIcon({ status }: { status: ToolCallPart['status'] }) {
if (status === 'running' || status === 'pending') {
return (
<Loader2 className="size-3.5 shrink-0 animate-spin text-muted-foreground" />
)
}
if (status === 'completed') {
return <CheckCircle2 className="size-3.5 shrink-0 text-green-500" />
}
return <XCircle className="size-3.5 shrink-0 text-destructive" />
}
interface ClawChatMessageProps {
message: ClawChatMessageType
}
export const ClawChatMessage: FC<ClawChatMessageProps> = ({ message }) => {
const messageText = message.parts
.filter((p) => p.type === 'text')
.map((p) => p.text)
.join('\n')
const handleCopy = useCallback(() => {
if (messageText) navigator.clipboard.writeText(messageText)
}, [messageText])
const entries = useMemo(
() => buildRenderEntries(message.parts),
[message.parts],
)
return (
<Message
from={message.role}
className="max-w-full group-[.is-user]:max-w-[80%]"
>
<MessageContent className="max-w-full overflow-hidden group-[.is-assistant]:w-full group-[.is-user]:max-w-full">
{entries.map((entry) => {
const key = `${message.id}-entry-${entry.partIndex}`
if (entry.kind === 'attachments' && entry.attachments) {
return (
<MessageAttachments key={key}>
{entry.attachments.map((attachment, idx) => (
<MessageAttachment
// biome-ignore lint/suspicious/noArrayIndexKey: attachment order is stable within a finalized message
key={`${attachment.kind}-${idx}`}
data={{
type: 'file',
url: attachment.dataUrl ?? '',
mediaType: attachment.mediaType,
filename: attachment.name,
}}
/>
))}
</MessageAttachments>
)
}
if (entry.kind === 'text' && entry.part?.type === 'text') {
return (
<MessageResponse
key={key}
// Historical messages are finalized — render immediately.
// Streamdown's default "streaming" mode uses an idle-callback
// debounce (300ms / 500ms idle) that paints empty content
// first, which made history flash blank tool collapsibles
// before text on every load.
mode="static"
parseIncompleteMarkdown={false}
className={cn(
'max-w-full overflow-hidden break-words',
'[&_[data-streamdown="code-block"]]:!w-full [&_[data-streamdown="code-block"]]:!max-w-full [&_[data-streamdown="code-block"]]:overflow-x-auto',
'[&_[data-streamdown="table-wrapper"]]:!w-full [&_[data-streamdown="table-wrapper"]]:!max-w-full [&_[data-streamdown="table-wrapper"]]:overflow-x-auto',
'[&_table]:w-max [&_table]:min-w-full',
)}
>
{entry.part.text}
</MessageResponse>
)
}
if (entry.kind === 'reasoning' && entry.part?.type === 'reasoning') {
return (
<Reasoning
key={key}
className="w-full"
defaultOpen={false}
duration={entry.part.duration}
>
<ReasoningTrigger />
<ReasoningContent>{entry.part.text}</ReasoningContent>
</Reasoning>
)
}
if (entry.kind === 'meta' && entry.part?.type === 'meta') {
return (
<div key={key} className="text-muted-foreground text-xs">
{entry.part.label}: {entry.part.value}
</div>
)
}
if (entry.kind === 'task' && entry.tools) {
const tools = entry.tools
const errorCount = tools.filter((t) => t.status === 'failed').length
const taskTitle = `Agent activity (${tools.length} ${tools.length === 1 ? 'action' : 'actions'}${errorCount > 0 ? `, ${errorCount} failed` : ''})`
return (
<Task key={key} defaultOpen={false}>
<TaskTrigger title={taskTitle} TriggerIcon={Wrench} />
<TaskContent>
{tools.map((tool, idx) => (
<TaskItem
// biome-ignore lint/suspicious/noArrayIndexKey: tool order is stable within a finalized historical message
key={`${tool.name}-${tool.status}-${idx}`}
className="flex items-center gap-2"
>
<ToolStatusIcon status={tool.status} />
<span className="text-foreground text-xs">
{tool.label}
</span>
{tool.subject ? (
<span className="ml-1.5 truncate text-muted-foreground/70 text-xs">
· {tool.subject}
</span>
) : null}
{tool.error ? (
<span className="ml-2 truncate text-destructive text-xs">
{tool.error}
</span>
) : null}
{tool.durationMs != null ? (
<span className="ml-auto text-muted-foreground/60 text-xs tabular-nums">
{(tool.durationMs / 1000).toFixed(1)}s
</span>
) : null}
</TaskItem>
))}
</TaskContent>
</Task>
)
}
return null
})}
{message.role === 'assistant' && messageText ? (
<MessageToolbar>
<MessageActions>
<MessageAction tooltip="Copy" onClick={handleCopy}>
<Copy className="size-3.5" />
</MessageAction>
</MessageActions>
{message.costUsd ? (
<span className="text-[11px] text-muted-foreground/50 tabular-nums">
{formatCost(message.costUsd)}
</span>
) : null}
</MessageToolbar>
) : null}
</MessageContent>
</Message>
)
}

View File

@@ -1,21 +1,36 @@
import {
AlertTriangle,
ArrowRight,
Bot,
ChevronDown,
FileText,
Folder,
Layers,
Loader2,
Mic,
Paperclip,
RefreshCw,
Square,
X,
} from 'lucide-react'
import { type FC, type ReactNode, useEffect, useState } from 'react'
import {
type DragEvent,
type FC,
type ReactNode,
useEffect,
useLayoutEffect,
useRef,
useState,
} from 'react'
import { AppSelector } from '@/components/elements/AppSelector'
import { TabPickerPopover } from '@/components/elements/tab-picker-popover'
import { WorkspaceSelector } from '@/components/elements/workspace-selector'
import { Button } from '@/components/ui/button'
import { Textarea } from '@/components/ui/textarea'
import type { AgentEntry } from '@/entrypoints/app/agents/useOpenClaw'
import { McpServerIcon } from '@/entrypoints/app/connect-mcp/McpServerIcon'
import { useGetUserMCPIntegrations } from '@/entrypoints/app/connect-mcp/useGetUserMCPIntegrations'
import { type StagedAttachment, stageAttachments } from '@/lib/attachments'
import { Feature } from '@/lib/browseros/capabilities'
import { useCapabilities } from '@/lib/browseros/useCapabilities'
import { useMcpServers } from '@/lib/mcp/mcpServerStorage'
@@ -23,18 +38,33 @@ import { cn } from '@/lib/utils'
import { useVoiceInput } from '@/lib/voice/useVoiceInput'
import { useWorkspace } from '@/lib/workspace/use-workspace'
import { AgentSelector } from './AgentSelector'
import type { OutboundMessage } from './useOutboundQueue'
export interface ConversationInputSendInput {
text: string
attachments: StagedAttachment[]
}
interface ConversationInputProps {
agents: AgentEntry[]
selectedAgentId: string | null
onSelectAgent: (agent: AgentEntry) => void
onSend: (text: string) => void
onSend: (input: ConversationInputSendInput) => void
onCreateAgent?: () => void
streaming: boolean
disabled?: boolean
status?: string
placeholder?: string
variant?: 'home' | 'conversation'
// Outbound queue: when present, the composer renders the queue strip
// above the textarea and lets the user keep sending while a previous
// turn is in flight. Optional so non-conversation variants (the home
// page) can opt out — the queue only makes sense in the conversation
// page where each enqueued message will eventually be delivered to the
// active agent.
outboundQueue?: OutboundMessage[]
onCancelQueued?: (id: string) => void
onRetryQueued?: (id: string) => void
}
function InputActionButton({
@@ -123,6 +153,8 @@ function ContextControls({
onToggleTab,
showAgentSelector,
status,
onAttachClick,
attachDisabled,
}: {
agents: AgentEntry[]
onCreateAgent?: () => void
@@ -132,6 +164,8 @@ function ContextControls({
onToggleTab: (tab: chrome.tabs.Tab) => void
showAgentSelector: boolean
status?: string
onAttachClick: () => void
attachDisabled: boolean
}) {
const { supports } = useCapabilities()
const { selectedFolder } = useWorkspace()
@@ -146,7 +180,7 @@ function ContextControls({
})
return (
<div className="flex items-center justify-between border-border/50 border-t px-5 py-3">
<div className="flex items-center justify-between border-border/40 border-t px-4 py-2.5">
<div className="flex items-center gap-1">
{showAgentSelector ? (
<AgentSelector
@@ -191,6 +225,20 @@ function ContextControls({
<span>Tabs</span>
</Button>
</TabPickerPopover>
<Button
type="button"
variant="ghost"
onClick={onAttachClick}
disabled={attachDisabled}
title="Attach files"
className={cn(
'flex items-center gap-2 rounded-lg px-3 py-1.5 font-medium text-sm transition-all',
'bg-transparent text-muted-foreground hover:bg-accent hover:text-accent-foreground',
)}
>
<Paperclip className="h-4 w-4" />
<span>Attach</span>
</Button>
</div>
{supports(Feature.MANAGED_MCP_SUPPORT) ? (
@@ -234,7 +282,7 @@ function ContextControls({
function HomeShell({ children }: { children: ReactNode }) {
return (
<div className="overflow-hidden rounded-[1.5rem] border border-border/60 bg-card/95 shadow-sm backdrop-blur">
<div className="overflow-hidden rounded-[1.55rem] border border-border/60 bg-card/95 shadow-sm">
{children}
</div>
)
@@ -242,7 +290,7 @@ function HomeShell({ children }: { children: ReactNode }) {
function ConversationShell({ children }: { children: ReactNode }) {
return (
<div className="overflow-hidden rounded-[1.5rem] border border-border/60 bg-card/95 shadow-sm backdrop-blur">
<div className="overflow-hidden rounded-[1.35rem] border border-border/50 bg-background/95 shadow-[0_10px_30px_rgba(15,23,42,0.06)] backdrop-blur-md">
{children}
</div>
)
@@ -259,13 +307,60 @@ export const ConversationInput: FC<ConversationInputProps> = ({
status,
placeholder,
variant = 'conversation',
outboundQueue,
onCancelQueued,
onRetryQueued,
}) => {
const [input, setInput] = useState('')
const [selectedTabs, setSelectedTabs] = useState<chrome.tabs.Tab[]>([])
const [isExpandedDraft, setIsExpandedDraft] = useState(false)
const [attachments, setAttachments] = useState<StagedAttachment[]>([])
const [attachmentError, setAttachmentError] = useState<string | null>(null)
const [isStaging, setIsStaging] = useState(false)
const [isDragOver, setIsDragOver] = useState(false)
const fileInputRef = useRef<HTMLInputElement>(null)
const voice = useVoiceInput()
const textareaRef = useRef<HTMLTextAreaElement>(null)
const selectedAgent = agents.find(
(agent) => agent.agentId === selectedAgentId,
)
const isConversation = variant === 'conversation'
const stageFiles = async (files: File[]) => {
if (files.length === 0) return
setIsStaging(true)
setAttachmentError(null)
try {
const result = await stageAttachments(files, attachments.length)
if (result.staged.length > 0) {
setAttachments((prev) => [...prev, ...result.staged])
}
if (result.errors.length > 0) {
setAttachmentError(result.errors.map((e) => e.message).join(' \u2022 '))
}
} finally {
setIsStaging(false)
}
}
const removeAttachment = (id: string) => {
setAttachments((prev) => prev.filter((a) => a.id !== id))
setAttachmentError(null)
}
useLayoutEffect(() => {
const element = textareaRef.current
if (!element) return
const maxHeight = isConversation ? 176 : 100
const collapsedHeight = isConversation ? 56 : 72
element.style.height = '0px'
const nextHeight = Math.min(element.scrollHeight, maxHeight)
element.style.height = `${nextHeight}px`
element.style.overflowY =
element.scrollHeight > maxHeight ? 'auto' : 'hidden'
setIsExpandedDraft(nextHeight > collapsedHeight)
})
useEffect(() => {
if (voice.transcript && !voice.isTranscribing) {
@@ -284,11 +379,71 @@ export const ConversationInput: FC<ConversationInputProps> = ({
})
}
const hasContent = input.trim().length > 0 || attachments.length > 0
const queueEnabled = outboundQueue !== undefined
const handleSend = () => {
const text = input.trim()
if (!text || streaming || disabled) return
onSend(text)
// The outbound queue accepts new messages while streaming; legacy
// direct-send callers (e.g., the home composer) keep the original
// streaming-blocks-send semantic.
if (disabled || isStaging) return
if (!queueEnabled && streaming) return
if (!text && attachments.length === 0) return
onSend({ text, attachments })
setInput('')
setAttachments([])
setAttachmentError(null)
}
const handlePaste = (event: React.ClipboardEvent<HTMLTextAreaElement>) => {
const items = event.clipboardData?.items
if (!items) return
const files: File[] = []
for (const item of items) {
if (item.kind === 'file') {
const file = item.getAsFile()
if (file) files.push(file)
}
}
if (files.length > 0) {
event.preventDefault()
void stageFiles(files)
}
}
const handleDrop = (event: DragEvent<HTMLDivElement>) => {
event.preventDefault()
setIsDragOver(false)
const files = Array.from(event.dataTransfer?.files ?? [])
if (files.length > 0) {
void stageFiles(files)
}
}
const handleDragOver = (event: DragEvent<HTMLDivElement>) => {
if (!event.dataTransfer?.types.includes('Files')) return
event.preventDefault()
setIsDragOver(true)
}
const handleDragLeave = (event: DragEvent<HTMLDivElement>) => {
if (event.currentTarget.contains(event.relatedTarget as Node | null)) {
return
}
setIsDragOver(false)
}
const openFilePicker = () => {
fileInputRef.current?.click()
}
const handleFileInputChange = (
event: React.ChangeEvent<HTMLInputElement>,
) => {
const files = Array.from(event.target.files ?? [])
event.target.value = ''
if (files.length > 0) void stageFiles(files)
}
const shell = variant === 'home' ? HomeShell : ConversationShell
@@ -296,73 +451,321 @@ export const ConversationInput: FC<ConversationInputProps> = ({
return (
<Shell>
<div className="flex items-center gap-3 px-5 py-4">
<BotInputIcon variant={variant} />
<section
// Drag/drop on a region isn't a click affordance — wrap the
// composer in a labeled <section> so the a11y rule is satisfied
// without misrepresenting the surface as interactive.
aria-label="Message composer"
className={cn('relative', isDragOver && 'ring-2 ring-primary/60')}
onDragOver={handleDragOver}
onDragLeave={handleDragLeave}
onDrop={handleDrop}
>
<input
type="text"
value={input}
onChange={(event) => setInput(event.currentTarget.value)}
onKeyDown={(event) => {
if (event.key === 'Enter') {
event.preventDefault()
handleSend()
ref={fileInputRef}
type="file"
multiple
accept="image/png,image/jpeg,image/webp,image/gif,text/*,application/json"
className="hidden"
onChange={handleFileInputChange}
/>
{attachments.length > 0 || attachmentError ? (
<AttachmentStrip
attachments={attachments}
onRemove={removeAttachment}
error={attachmentError}
/>
) : null}
{queueEnabled && outboundQueue && outboundQueue.length > 0 ? (
<OutboundQueueStrip
messages={outboundQueue}
onCancel={onCancelQueued}
onRetry={onRetryQueued}
/>
) : null}
<div
className={cn(
'flex gap-3',
variant === 'home' ? 'px-4 py-3' : 'px-4 py-3',
isExpandedDraft ? 'items-end' : 'items-center',
)}
>
<BotInputIcon variant={variant} />
<div className="flex-1">
<Textarea
ref={textareaRef}
value={input}
onChange={(event) => setInput(event.currentTarget.value)}
onKeyDown={(event) => {
if (event.key === 'Enter' && !event.shiftKey) {
event.preventDefault()
handleSend()
}
}}
onPaste={handlePaste}
rows={1}
placeholder={
voice.isTranscribing
? 'Transcribing...'
: (placeholder ??
`Message ${selectedAgent?.name ?? 'agent'}...`)
}
disabled={disabled || voice.isTranscribing}
className={cn(
'resize-none border-none bg-transparent px-0 text-[15px] shadow-none focus-visible:ring-0',
'[field-sizing:fixed]',
variant === 'home'
? 'min-h-[40px] py-2 leading-6'
: 'min-h-[40px] py-2 leading-6',
'placeholder:text-muted-foreground/80',
)}
/>
</div>
<VoiceButton
isRecording={voice.isRecording}
isTranscribing={voice.isTranscribing}
onStart={() => {
void voice.startRecording()
}}
onStop={() => {
void voice.stopRecording()
}}
/>
<InputActionButton
disabled={
!hasContent ||
isStaging ||
!!disabled ||
voice.isRecording ||
voice.isTranscribing ||
// Only block on `streaming` for the legacy direct-send path
// (no queue). With the queue active the press always
// succeeds — it just enqueues instead of dispatching.
(!queueEnabled && streaming)
}
}}
placeholder={
voice.isTranscribing
? 'Transcribing...'
: (placeholder ?? `Message ${selectedAgent?.name ?? 'agent'}...`)
}
disabled={disabled || voice.isTranscribing}
className="flex-1 border-none bg-transparent text-base text-foreground outline-none placeholder:text-muted-foreground disabled:opacity-60"
onClick={handleSend}
// Spinner stays the user-facing "agent is busy" hint; with the
// queue active we still spin while a turn is in flight.
streaming={streaming}
/>
</div>
{voice.error ? (
<div className="px-5 pb-2 text-destructive text-xs">
{voice.error}
</div>
) : null}
<ContextControls
agents={agents}
onCreateAgent={onCreateAgent}
onSelectAgent={onSelectAgent}
selectedAgentId={selectedAgentId}
selectedTabs={selectedTabs}
onToggleTab={toggleTab}
showAgentSelector={variant === 'home'}
status={status}
onAttachClick={openFilePicker}
attachDisabled={attachments.length >= 10 || isStaging || !!disabled}
/>
<VoiceButton
isRecording={voice.isRecording}
isTranscribing={voice.isTranscribing}
onStart={() => {
void voice.startRecording()
}}
onStop={() => {
void voice.stopRecording()
}}
/>
<InputActionButton
disabled={
!input.trim() ||
streaming ||
!!disabled ||
voice.isRecording ||
voice.isTranscribing
}
onClick={handleSend}
streaming={streaming}
/>
</div>
{voice.error ? (
<div className="px-5 pb-2 text-destructive text-xs">{voice.error}</div>
) : null}
<ContextControls
agents={agents}
onCreateAgent={onCreateAgent}
onSelectAgent={onSelectAgent}
selectedAgentId={selectedAgentId}
selectedTabs={selectedTabs}
onToggleTab={toggleTab}
showAgentSelector={variant === 'home'}
status={status}
/>
{isDragOver ? (
<div className="pointer-events-none absolute inset-0 flex items-center justify-center rounded-[inherit] bg-background/80 font-medium text-foreground text-sm backdrop-blur-sm">
Drop files to attach
</div>
) : null}
</section>
</Shell>
)
}
function OutboundQueueStrip({
messages,
onCancel,
onRetry,
}: {
messages: OutboundMessage[]
onCancel?: (id: string) => void
onRetry?: (id: string) => void
}) {
return (
<div className="border-border/40 border-b px-4 pt-3 pb-2">
<ul className="flex flex-col gap-1">
{messages.map((message) => (
<OutboundQueueItem
key={message.id}
message={message}
onCancel={onCancel}
onRetry={onRetry}
/>
))}
</ul>
</div>
)
}
function OutboundQueueItem({
message,
onCancel,
onRetry,
}: {
message: OutboundMessage
onCancel?: (id: string) => void
onRetry?: (id: string) => void
}) {
const preview = message.text.trim() || '(attachments only)'
return (
<li className="flex items-center gap-2 rounded-md px-2 py-1 text-xs">
<OutboundQueueStatusIcon status={message.status} />
<span className="min-w-0 flex-1 truncate text-muted-foreground">
{preview}
</span>
{message.attachmentPreviews.length > 0 ? (
<span className="inline-flex items-center gap-1 text-muted-foreground/70">
<Paperclip className="size-3" />
<span className="tabular-nums">
{message.attachmentPreviews.length}
</span>
</span>
) : null}
{message.status === 'queued' && onCancel ? (
<button
type="button"
onClick={() => onCancel(message.id)}
className="ml-1 inline-flex size-5 items-center justify-center rounded-full text-muted-foreground hover:bg-accent hover:text-foreground"
aria-label="Cancel queued message"
title="Cancel"
>
<X className="size-3" />
</button>
) : null}
{message.status === 'failed' ? (
<span className="ml-1 inline-flex items-center gap-2 text-destructive">
<span className="max-w-[160px] truncate" title={message.error}>
{message.error ?? 'Failed'}
</span>
{onRetry ? (
<button
type="button"
onClick={() => onRetry(message.id)}
className="inline-flex size-5 items-center justify-center rounded-full hover:bg-accent hover:text-foreground"
aria-label="Retry failed message"
title="Retry"
>
<RefreshCw className="size-3" />
</button>
) : null}
{onCancel ? (
<button
type="button"
onClick={() => onCancel(message.id)}
className="inline-flex size-5 items-center justify-center rounded-full hover:bg-accent hover:text-foreground"
aria-label="Discard failed message"
title="Discard"
>
<X className="size-3" />
</button>
) : null}
</span>
) : null}
</li>
)
}
function OutboundQueueStatusIcon({
status,
}: {
status: OutboundMessage['status']
}) {
if (status === 'sending') {
return (
<Loader2 className="size-3.5 shrink-0 animate-spin text-muted-foreground" />
)
}
if (status === 'failed') {
return <AlertTriangle className="size-3.5 shrink-0 text-destructive" />
}
return (
<span className="inline-block size-2 shrink-0 rounded-full bg-muted-foreground/40" />
)
}
function AttachmentStrip({
attachments,
onRemove,
error,
}: {
attachments: StagedAttachment[]
onRemove: (id: string) => void
error: string | null
}) {
return (
<div className="border-border/40 border-b px-4 pt-3 pb-2">
{attachments.length > 0 ? (
<div className="flex flex-wrap gap-2">
{attachments.map((attachment) => (
<AttachmentChip
key={attachment.id}
attachment={attachment}
onRemove={() => onRemove(attachment.id)}
/>
))}
</div>
) : null}
{error ? (
<div className="mt-2 text-destructive text-xs">{error}</div>
) : null}
</div>
)
}
function AttachmentChip({
attachment,
onRemove,
}: {
attachment: StagedAttachment
onRemove: () => void
}) {
if (attachment.kind === 'image' && attachment.dataUrl) {
return (
<div className="group relative size-16 overflow-hidden rounded-md border border-border/60">
<img
src={attachment.dataUrl}
alt={attachment.name}
className="size-full object-cover"
/>
<button
type="button"
onClick={onRemove}
className="absolute top-1 right-1 inline-flex size-5 items-center justify-center rounded-full bg-background/80 text-muted-foreground opacity-0 transition-opacity hover:text-foreground group-hover:opacity-100"
aria-label={`Remove ${attachment.name}`}
>
<X className="size-3" />
</button>
</div>
)
}
return (
<div className="group flex max-w-[220px] items-center gap-2 rounded-md border border-border/60 bg-background/60 px-2 py-1.5">
<FileText className="size-4 shrink-0 text-muted-foreground" />
<span className="truncate text-xs">{attachment.name}</span>
<button
type="button"
onClick={onRemove}
className="ml-1 inline-flex size-4 items-center justify-center text-muted-foreground hover:text-foreground"
aria-label={`Remove ${attachment.name}`}
>
<X className="size-3" />
</button>
</div>
)
}
function BotInputIcon({ variant }: { variant: 'home' | 'conversation' }) {
return (
<div
className={cn(
'flex items-center justify-center text-[var(--accent-orange)]',
variant === 'home'
? 'h-10 w-10 rounded-xl bg-[var(--accent-orange)]/10'
: 'h-9 w-9 rounded-xl bg-[var(--accent-orange)]/12',
? 'h-8 w-8 rounded-lg bg-[var(--accent-orange)]/10'
: 'h-8 w-8 rounded-lg bg-[var(--accent-orange)]/10',
)}
>
<Bot className="h-4 w-4" />

View File

@@ -1,7 +1,9 @@
import { Bot, CheckCircle2, Loader2, XCircle } from 'lucide-react'
import type { FC } from 'react'
import { Bot, CheckCircle2, Loader2, Wrench, XCircle } from 'lucide-react'
import { type FC, useMemo } from 'react'
import {
Message,
MessageAttachment,
MessageAttachments,
MessageContent,
MessageResponse,
} from '@/components/ai-elements/message'
@@ -10,96 +12,191 @@ import {
ReasoningContent,
ReasoningTrigger,
} from '@/components/ai-elements/reasoning'
import type { AgentConversationTurn } from '@/lib/agent-conversations/types'
import {
Task,
TaskContent,
TaskItem,
TaskTrigger,
} from '@/components/ai-elements/task'
import type {
AgentConversationTurn,
ToolEntry,
} from '@/lib/agent-conversations/types'
interface ConversationMessageProps {
turn: AgentConversationTurn
streaming: boolean
}
interface RenderEntry {
kind: 'thinking' | 'text' | 'task'
partIndex: number
text?: string
done?: boolean
tools?: ToolEntry[]
}
/**
* Build the render plan for an assistant turn:
* - thinking and text parts render in place
* - all tool-batch parts collapse into a single Task entry at their first
* appearance position, with tools listed in arrival order
*/
function buildRenderEntries(turn: AgentConversationTurn): RenderEntry[] {
const entries: RenderEntry[] = []
const aggregatedTools: ToolEntry[] = []
let taskInserted = false
turn.parts.forEach((part, partIndex) => {
if (part.kind === 'thinking') {
entries.push({
kind: 'thinking',
partIndex,
text: part.text,
done: part.done,
})
} else if (part.kind === 'text') {
entries.push({ kind: 'text', partIndex, text: part.text })
} else if (part.kind === 'tool-batch') {
aggregatedTools.push(...part.tools)
if (!taskInserted) {
entries.push({
kind: 'task',
partIndex,
tools: aggregatedTools,
})
taskInserted = true
}
}
})
return entries
}
function ToolStatusIcon({ status }: { status: ToolEntry['status'] }) {
if (status === 'running') {
return (
<Loader2 className="size-3.5 shrink-0 animate-spin text-muted-foreground" />
)
}
if (status === 'completed') {
return <CheckCircle2 className="size-3.5 shrink-0 text-green-500" />
}
return <XCircle className="size-3.5 shrink-0 text-destructive" />
}
export const ConversationMessage: FC<ConversationMessageProps> = ({
turn,
streaming,
}) => (
<div className="space-y-3">
<Message from="user">
<MessageContent>
<pre className="whitespace-pre-wrap font-sans text-sm">
{turn.userText}
</pre>
</MessageContent>
</Message>
}) => {
const entries = useMemo(() => buildRenderEntries(turn), [turn])
{turn.parts.length > 0 && (
<Message from="assistant">
return (
<div className="space-y-3">
<Message from="user">
<MessageContent>
{turn.parts.map((part, i) => {
const key = `${turn.id}-part-${i}`
{turn.userAttachments && turn.userAttachments.length > 0 && (
<MessageAttachments>
{turn.userAttachments.map((attachment) => (
<MessageAttachment
key={attachment.id}
data={{
type: 'file',
url: attachment.dataUrl ?? '',
mediaType: attachment.mediaType,
filename: attachment.name,
}}
/>
))}
</MessageAttachments>
)}
{turn.userText && (
<pre className="whitespace-pre-wrap font-sans text-sm">
{turn.userText}
</pre>
)}
</MessageContent>
</Message>
switch (part.kind) {
case 'thinking':
{entries.length > 0 && (
<Message from="assistant">
<MessageContent>
{entries.map((entry) => {
const key = `${turn.id}-entry-${entry.partIndex}`
if (entry.kind === 'thinking') {
return (
<Reasoning
key={key}
className="w-full"
isStreaming={!part.done}
defaultOpen={!part.done}
isStreaming={!entry.done}
defaultOpen={!entry.done}
>
<ReasoningTrigger />
<ReasoningContent>{part.text}</ReasoningContent>
<ReasoningContent>{entry.text ?? ''}</ReasoningContent>
</Reasoning>
)
}
case 'tool-batch':
if (entry.kind === 'text') {
return (
<div key={key} className="w-full space-y-1">
{part.tools.map((tool) => (
<div
<MessageResponse key={key}>
{entry.text ?? ''}
</MessageResponse>
)
}
const tools = entry.tools ?? []
const allDone = tools.every((t) => t.status !== 'running')
const taskTitle = allDone
? `Agent activity (${tools.length} ${tools.length === 1 ? 'action' : 'actions'})`
: `Working… (${tools.length} ${tools.length === 1 ? 'action' : 'actions'})`
return (
<Task key={key} defaultOpen={!turn.done}>
<TaskTrigger title={taskTitle} TriggerIcon={Wrench} />
<TaskContent>
{tools.map((tool) => (
<TaskItem
key={tool.id}
className="flex items-center gap-2 rounded-md border px-3 py-2 text-sm"
className="flex items-center gap-2"
>
{tool.status === 'running' && (
<Loader2 className="size-3.5 animate-spin text-muted-foreground" />
)}
{tool.status === 'completed' && (
<CheckCircle2 className="size-3.5 text-green-500" />
)}
{tool.status === 'error' && (
<XCircle className="size-3.5 text-destructive" />
)}
<span className="font-mono text-xs">{tool.name}</span>
<ToolStatusIcon status={tool.status} />
<span className="text-foreground text-xs">
{tool.label}
</span>
{tool.subject ? (
<span className="ml-1.5 truncate text-muted-foreground/70 text-xs">
· {tool.subject}
</span>
) : null}
{tool.durationMs != null && (
<span className="ml-auto text-muted-foreground text-xs">
<span className="ml-auto text-muted-foreground/60 text-xs tabular-nums">
{(tool.durationMs / 1000).toFixed(1)}s
</span>
)}
</div>
</TaskItem>
))}
</div>
)
</TaskContent>
</Task>
)
})}
</MessageContent>
</Message>
)}
case 'text':
return <MessageResponse key={key}>{part.text}</MessageResponse>
default:
return null
}
})}
</MessageContent>
</Message>
)}
{!turn.done && turn.parts.length === 0 && streaming && (
<div className="flex gap-2">
<div className="flex size-7 shrink-0 items-center justify-center rounded-full bg-[var(--accent-orange)] text-white">
<Bot className="size-3.5" />
{!turn.done && turn.parts.length === 0 && streaming && (
<div className="flex gap-2">
<div className="flex size-7 shrink-0 items-center justify-center rounded-full bg-[var(--accent-orange)] text-white">
<Bot className="size-3.5" />
</div>
<div className="flex items-center gap-1 rounded-xl rounded-tl-none border border-border/50 bg-card px-3 py-2.5 shadow-sm">
<span className="size-1.5 animate-bounce rounded-full bg-[var(--accent-orange)] [animation-delay:-0.3s]" />
<span className="size-1.5 animate-bounce rounded-full bg-[var(--accent-orange)] [animation-delay:-0.15s]" />
<span className="size-1.5 animate-bounce rounded-full bg-[var(--accent-orange)]" />
</div>
</div>
<div className="flex items-center gap-1 rounded-xl rounded-tl-none border border-border/50 bg-card px-3 py-2.5 shadow-sm">
<span className="size-1.5 animate-bounce rounded-full bg-[var(--accent-orange)] [animation-delay:-0.3s]" />
<span className="size-1.5 animate-bounce rounded-full bg-[var(--accent-orange)] [animation-delay:-0.15s]" />
<span className="size-1.5 animate-bounce rounded-full bg-[var(--accent-orange)]" />
</div>
</div>
)}
</div>
)
)}
</div>
)
}

View File

@@ -0,0 +1,121 @@
import { describe, expect, it } from 'bun:test'
import {
type AgentHistoryPageResponse,
type BrowserOSChatHistoryItem,
buildChatHistoryFromClawMessages,
flattenHistoryPages,
mapHistoryItemToClawMessage,
} from './claw-chat-types'
function historyItem(
overrides: Partial<BrowserOSChatHistoryItem>,
): BrowserOSChatHistoryItem {
return {
id: 'session-1:0',
role: 'user',
text: 'Hello',
timestamp: 1000,
messageSeq: 0,
sessionKey: 'session-1',
source: 'user-chat',
...overrides,
}
}
function page(items: BrowserOSChatHistoryItem[]): AgentHistoryPageResponse {
return {
agentId: 'main',
sessionKey: 'session-1',
session: null,
items,
page: {
hasMore: false,
limit: 50,
},
}
}
describe('claw-chat-types', () => {
it('maps backend history items into text-first ClawChat messages', () => {
const message = mapHistoryItemToClawMessage(
historyItem({
id: 'session-1:1',
role: 'assistant',
text: 'Hi there',
messageSeq: 1,
}),
)
expect(message).toEqual({
id: 'session-1:1',
role: 'assistant',
sessionKey: 'session-1',
timestamp: 1000,
source: 'user-chat',
messageSeq: 1,
status: 'historical',
parts: [{ type: 'text', text: 'Hi there' }],
})
})
it('flattens paginated history into oldest-to-newest render order', () => {
const messages = flattenHistoryPages([
page([
historyItem({
id: 'session-1:2',
role: 'user',
text: 'newer',
timestamp: 3000,
messageSeq: 2,
}),
]),
page([
historyItem({
id: 'session-1:0',
role: 'user',
text: 'older',
timestamp: 1000,
messageSeq: 0,
}),
historyItem({
id: 'session-1:1',
role: 'assistant',
text: 'middle',
timestamp: 2000,
messageSeq: 1,
}),
]),
])
expect(messages.map((message) => message.id)).toEqual([
'session-1:0',
'session-1:1',
'session-1:2',
])
})
it('builds OpenClaw chat history from text message parts only', () => {
const history = buildChatHistoryFromClawMessages([
{
id: 'user-1',
role: 'user',
sessionKey: 'session-1',
parts: [{ type: 'text', text: ' User request ' }],
},
{
id: 'assistant-1',
role: 'assistant',
sessionKey: 'session-1',
parts: [
{ type: 'reasoning', text: 'private reasoning' },
{ type: 'text', text: 'Assistant answer' },
],
},
])
expect(history).toEqual([
{ role: 'user', content: 'User request' },
{ role: 'assistant', content: 'Assistant answer' },
])
})
})

View File

@@ -0,0 +1,223 @@
import type { OpenClawChatHistoryMessage } from '@/entrypoints/app/agents/useOpenClaw'
export type ClawChatRole = 'user' | 'assistant'
export type ClawChatSource = 'user-chat' | 'cron' | 'hook' | 'channel' | 'other'
export interface BrowserOSOpenClawSession {
key: string
updatedAt: number
sessionId: string
agentId: string
kind: string
source: ClawChatSource
status?: string
totalTokens?: number
model?: string
modelProvider?: string
}
export interface BrowserOSChatHistoryToolCall {
toolCallId?: string
toolName: string
label: string
subject?: string
status: 'completed' | 'failed'
input?: Record<string, unknown>
output?: string
error?: string
durationMs?: number
}
export interface BrowserOSChatHistoryReasoning {
text: string
durationMs?: number
}
export interface BrowserOSChatHistoryAttachment {
kind: 'image' | 'file'
mediaType: string
// Images carry a `data:` URL so we can render directly without any
// additional fetch; files (text/PDF) currently round-trip via inline
// text in the message body and do not populate this field in v1.
dataUrl?: string
name?: string
}
export interface BrowserOSChatHistoryItem {
id: string
role: ClawChatRole
text: string
timestamp?: number
messageSeq: number
sessionKey: string
source: ClawChatSource
costUsd?: number
tokensIn?: number
tokensOut?: number
toolCalls?: BrowserOSChatHistoryToolCall[]
reasoning?: BrowserOSChatHistoryReasoning
attachments?: BrowserOSChatHistoryAttachment[]
}
export interface AgentHistoryPageResponse {
agentId: string
sessionKey: string | null
session: BrowserOSOpenClawSession | null
items: BrowserOSChatHistoryItem[]
page: {
cursor?: string
hasMore: boolean
limit: number
}
}
export type ClawChatMessageStatus =
| 'historical'
| 'sending'
| 'streaming'
| 'error'
export type ClawChatMessagePart =
| { type: 'text'; text: string }
| { type: 'reasoning'; text: string; duration?: number }
| {
type: 'tool-call'
name: string
label: string
subject?: string
status: 'pending' | 'running' | 'completed' | 'failed'
input?: unknown
output?: unknown
error?: string
durationMs?: number
}
| {
type: 'attachment'
kind: 'image' | 'file'
mediaType: string
dataUrl?: string
name?: string
}
| { type: 'meta'; label: string; value: string }
export interface ClawChatMessage {
id: string
role: ClawChatRole
sessionKey: string
timestamp?: number
source?: ClawChatSource
messageSeq?: number
status?: ClawChatMessageStatus
parts: ClawChatMessagePart[]
costUsd?: number
tokensIn?: number
tokensOut?: number
}
export function mapHistoryItemToClawMessage(
item: BrowserOSChatHistoryItem,
): ClawChatMessage {
const parts: ClawChatMessagePart[] = []
// Attachments first — they belong above the text in user messages and
// never appear on assistant messages today (assistant images come back
// through tool results, which render via the Task collapsible).
if (item.attachments && item.attachments.length > 0) {
for (const attachment of item.attachments) {
parts.push({
type: 'attachment',
kind: attachment.kind,
mediaType: attachment.mediaType,
dataUrl: attachment.dataUrl,
name: attachment.name,
})
}
}
// Reasoning, then tool calls, then text — the chronological order the
// agent produced them (think → act → answer).
if (item.reasoning && item.reasoning.text.trim().length > 0) {
// 0ms means thinking and the final answer were emitted in the same JSONL
// line (no tool calls between them) — there's no real elapsed wall-clock,
// so fall through to the "Thinking" trigger instead of "Thought for 0
// seconds" / streaming shimmer. Real multi-line turns floor at 1s.
const durationMs = item.reasoning.durationMs ?? 0
const duration =
durationMs > 0 ? Math.max(1, Math.round(durationMs / 1000)) : undefined
parts.push({
type: 'reasoning',
text: item.reasoning.text,
duration,
})
}
if (item.toolCalls && item.toolCalls.length > 0) {
for (const tc of item.toolCalls) {
parts.push({
type: 'tool-call',
name: tc.toolName,
label: tc.label,
subject: tc.subject,
status: tc.status,
input: tc.input,
output: tc.output,
error: tc.error,
durationMs: tc.durationMs,
})
}
}
// Only emit a text part when there's actual content. User messages with
// only attachments and no caption shouldn't render an empty bubble.
if (item.text.trim().length > 0) {
parts.push({ type: 'text', text: item.text })
}
return {
id: item.id,
role: item.role,
sessionKey: item.sessionKey,
timestamp: item.timestamp,
source: item.source,
messageSeq: item.messageSeq,
status: 'historical',
parts,
costUsd: item.costUsd,
tokensIn: item.tokensIn,
tokensOut: item.tokensOut,
}
}
export function flattenHistoryPages(
pages: AgentHistoryPageResponse[],
): ClawChatMessage[] {
return pages
.flatMap((page) => page.items)
.sort((a, b) => {
if (a.timestamp != null && b.timestamp != null) {
return a.timestamp - b.timestamp
}
return a.messageSeq - b.messageSeq
})
.map(mapHistoryItemToClawMessage)
}
export function buildChatHistoryFromClawMessages(
messages: ClawChatMessage[],
): OpenClawChatHistoryMessage[] {
return messages
.map((message) => {
const content = message.parts
.filter((part): part is { type: 'text'; text: string } => {
return part.type === 'text' && part.text.trim().length > 0
})
.map((part) => part.text.trim())
.join('\n\n')
return content ? { role: message.role, content } : null
})
.filter((message): message is OpenClawChatHistoryMessage =>
Boolean(message),
)
}

View File

@@ -1,69 +1,50 @@
import { useEffect, useState } from 'react'
import {
type AgentEntry,
getModelDisplayName,
type OpenClawStatus,
} from '@/entrypoints/app/agents/useOpenClaw'
import { getLatestConversation } from '@/lib/agent-conversations/storage'
import type { AgentCardData } from '@/lib/agent-conversations/types'
import type { AgentOverview } from './useAgentDashboard'
function getAgentStatusTone(
status: OpenClawStatus['status'] | undefined,
function resolveAgentStatus(
gatewayStatus: OpenClawStatus['status'] | undefined,
liveStatus: AgentOverview['status'] | undefined,
): AgentCardData['status'] {
if (status === 'error') return 'error'
if (status === 'starting') return 'working'
// Gateway-level errors take precedence
if (gatewayStatus === 'error') return 'error'
if (gatewayStatus === 'starting') return 'working'
// Per-agent live status from the WS observer
if (liveStatus === 'working') return 'working'
if (liveStatus === 'error') return 'error'
return 'idle'
}
async function getAgentCardData(
agent: AgentEntry,
status: OpenClawStatus['status'] | undefined,
): Promise<AgentCardData> {
const conversation = await getLatestConversation(agent.agentId)
const lastTurn = conversation?.turns[conversation.turns.length - 1]
const lastTextPart = lastTurn?.parts.findLast((part) => part.kind === 'text')
return {
agentId: agent.agentId,
name: agent.name,
model: getModelDisplayName(agent.model),
status: getAgentStatusTone(status),
lastMessage:
lastTextPart?.kind === 'text'
? lastTextPart.text.slice(0, 120)
: undefined,
lastMessageTimestamp: lastTurn?.timestamp,
}
}
export function useAgentCardData(
/**
* Build agent card display data by merging the raw agent entries from
* the gateway with enriched overview data from the dashboard API.
*
* Pure function — no hooks, no IndexedDB, no async.
*/
export function buildAgentCardData(
agents: AgentEntry[],
status: OpenClawStatus['status'] | undefined,
) {
const [cardData, setCardData] = useState<AgentCardData[]>([])
dashboard: AgentOverview[] | undefined,
): AgentCardData[] {
return agents.map((agent) => {
const overview = dashboard?.find((d) => d.agentId === agent.agentId)
useEffect(() => {
let active = true
const loadCardData = async () => {
const nextCardData = await Promise.all(
agents.map((agent) => getAgentCardData(agent, status)),
)
if (active) {
setCardData(nextCardData)
}
return {
agentId: agent.agentId,
name: agent.name,
model: getModelDisplayName(agent.model),
status: resolveAgentStatus(status, overview?.status),
lastMessage: overview?.latestMessage?.slice(0, 200) ?? undefined,
lastMessageTimestamp: overview?.latestMessageAt ?? undefined,
activitySummary: overview?.activitySummary ?? undefined,
currentTool: overview?.currentTool ?? undefined,
costUsd: overview?.totalCostUsd ?? undefined,
}
if (agents.length > 0) {
void loadCardData()
} else {
setCardData([])
}
return () => {
active = false
}
}, [agents, status])
return cardData
})
}

View File

@@ -1,51 +1,57 @@
import { useEffect, useRef, useState } from 'react'
import {
chatWithAgent,
type OpenClawChatHistoryMessage,
type OpenClawStreamEvent,
} from '@/entrypoints/app/agents/useOpenClaw'
import {
getLatestConversation,
saveConversation,
} from '@/lib/agent-conversations/storage'
import type {
AgentConversation,
AgentConversationTurn,
AssistantPart,
UserAttachmentPreview,
} from '@/lib/agent-conversations/types'
import type { ServerAttachmentPayload } from '@/lib/attachments'
import { consumeSSEStream } from '@/lib/sse'
import { buildToolLabel } from '@/lib/tool-labels'
export function useAgentConversation(agentId: string, agentName: string) {
export interface SendInput {
text: string
attachments?: ServerAttachmentPayload[]
// Optional preview metadata used to render the optimistic user turn.
// Built by the composer at staging time; the server only sees the
// payload array.
attachmentPreviews?: UserAttachmentPreview[]
}
interface UseAgentConversationOptions {
sessionKey?: string | null
history?: OpenClawChatHistoryMessage[]
onSessionKeyChange?: (sessionKey: string) => void
}
export function useAgentConversation(
agentId: string,
options: UseAgentConversationOptions = {},
) {
const [turns, setTurns] = useState<AgentConversationTurn[]>([])
const [streaming, setStreaming] = useState(false)
const [loading, setLoading] = useState(true)
const sessionKeyRef = useRef('')
const sessionKeyRef = useRef(options.sessionKey ?? '')
const historyRef = useRef<OpenClawChatHistoryMessage[]>(options.history ?? [])
const textAccRef = useRef('')
const thinkAccRef = useRef('')
const streamAbortRef = useRef<AbortController | null>(null)
const onSessionKeyChangeRef = useRef(options.onSessionKeyChange)
useEffect(() => {
let active = true
getLatestConversation(agentId)
.then((conv) => {
if (!active) return
if (conv) {
setTurns(conv.turns)
sessionKeyRef.current = conv.sessionKey
} else {
sessionKeyRef.current = crypto.randomUUID()
}
setLoading(false)
})
.catch(() => {
if (active) {
sessionKeyRef.current = crypto.randomUUID()
setLoading(false)
}
})
return () => {
active = false
}
}, [agentId])
sessionKeyRef.current = options.sessionKey ?? ''
}, [options.sessionKey])
useEffect(() => {
historyRef.current = options.history ?? []
}, [options.history])
useEffect(() => {
onSessionKeyChangeRef.current = options.onSessionKeyChange
}, [options.onSessionKeyChange])
useEffect(() => {
return () => {
@@ -53,18 +59,6 @@ export function useAgentConversation(agentId: string, agentName: string) {
}
}, [])
const persistTurns = (updatedTurns: AgentConversationTurn[]) => {
const conv: AgentConversation = {
agentId,
agentName,
sessionKey: sessionKeyRef.current,
turns: updatedTurns,
createdAt: updatedTurns[0]?.timestamp ?? Date.now(),
updatedAt: Date.now(),
}
saveConversation(conv).catch(() => {})
}
const updateCurrentTurnParts = (
updater: (parts: AssistantPart[]) => AssistantPart[],
) => {
@@ -110,9 +104,14 @@ export function useAgentConversation(agentId: string, agentName: string) {
}
case 'tool-start': {
const rawName = (event.data.toolName as string) ?? 'unknown'
const args = event.data.args as Record<string, unknown> | undefined
const { label, subject } = buildToolLabel(rawName, args)
const tool = {
id: (event.data.toolCallId as string) ?? crypto.randomUUID(),
name: (event.data.toolName as string) ?? 'unknown',
name: rawName,
label,
subject,
status: 'running' as const,
}
updateCurrentTurnParts((parts) => {
@@ -164,9 +163,7 @@ export function useAgentConversation(agentId: string, agentName: string) {
setTurns((prev) => {
const last = prev[prev.length - 1]
if (!last) return prev
const updated = [...prev.slice(0, -1), { ...last, done: true }]
persistTurns(updated)
return updated
return [...prev.slice(0, -1), { ...last, done: true }]
})
break
}
@@ -185,12 +182,22 @@ export function useAgentConversation(agentId: string, agentName: string) {
}
}
const send = async (text: string) => {
if (!text.trim() || streaming) return
const send = async (input: string | SendInput) => {
const normalized: SendInput =
typeof input === 'string' ? { text: input } : input
const trimmed = normalized.text.trim()
const attachments = normalized.attachments ?? []
if (streaming) return
if (!trimmed && attachments.length === 0) return
const turn: AgentConversationTurn = {
id: crypto.randomUUID(),
userText: text.trim(),
userText: trimmed,
userAttachments:
normalized.attachmentPreviews &&
normalized.attachmentPreviews.length > 0
? normalized.attachmentPreviews
: undefined,
parts: [],
done: false,
timestamp: Date.now(),
@@ -205,10 +212,17 @@ export function useAgentConversation(agentId: string, agentName: string) {
try {
const response = await chatWithAgent(
agentId,
text.trim(),
sessionKeyRef.current,
trimmed,
sessionKeyRef.current || undefined,
historyRef.current,
abortController.signal,
attachments,
)
const responseSessionKey = response.headers.get('X-Session-Key')
if (responseSessionKey) {
sessionKeyRef.current = responseSessionKey
onSessionKeyChangeRef.current?.(responseSessionKey)
}
if (!response.ok) {
const err = await response.text()
updateCurrentTurnParts((parts) => [
@@ -242,13 +256,11 @@ export function useAgentConversation(agentId: string, agentName: string) {
streamAbortRef.current = null
setTurns([])
setStreaming(false)
sessionKeyRef.current = crypto.randomUUID()
}
return {
turns,
streaming,
loading,
sessionKey: sessionKeyRef.current,
send,
resetConversation,

View File

@@ -0,0 +1,95 @@
import { useQuery, useQueryClient } from '@tanstack/react-query'
import { useEffect } from 'react'
import { useAgentServerUrl } from '@/lib/browseros/useBrowserOSProviders'
export interface AgentOverview {
agentId: string
status: 'working' | 'idle' | 'error' | 'unknown'
latestMessage: string | null
latestMessageAt: number | null
activitySummary: string | null
currentTool: string | null
totalCostUsd: number
sessionCount: number
}
export interface DashboardResponse {
agents: AgentOverview[]
summary: {
totalAgents: number
totalCostUsd: number
}
}
interface StatusEvent {
agentId: string
status: AgentOverview['status']
currentTool: string | null
error: string | null
timestamp: number
}
const DASHBOARD_QUERY_KEY = ['claw', 'dashboard']
export function useAgentDashboard(enabled: boolean) {
const { baseUrl, isLoading: urlLoading } = useAgentServerUrl()
const queryClient = useQueryClient()
const ready = enabled && Boolean(baseUrl) && !urlLoading
// Initial data load + periodic refresh as fallback
const query = useQuery<DashboardResponse>({
queryKey: [...DASHBOARD_QUERY_KEY, baseUrl],
queryFn: async () => {
const url = new URL('/claw/dashboard', baseUrl as string)
const response = await fetch(url.toString())
if (!response.ok) throw new Error('Failed to fetch dashboard')
return response.json()
},
enabled: ready,
})
// SSE subscription for real-time status patches
useEffect(() => {
if (!ready || !baseUrl) return
const streamUrl = new URL('/claw/dashboard/stream', baseUrl)
const eventSource = new EventSource(streamUrl.toString())
eventSource.addEventListener('snapshot', (event) => {
try {
const dashboard = JSON.parse(event.data) as DashboardResponse
queryClient.setQueryData([...DASHBOARD_QUERY_KEY, baseUrl], dashboard)
} catch {}
})
eventSource.addEventListener('status', (event) => {
try {
const status = JSON.parse(event.data) as StatusEvent
queryClient.setQueryData<DashboardResponse>(
[...DASHBOARD_QUERY_KEY, baseUrl],
(prev) => {
if (!prev) return prev
return {
...prev,
agents: prev.agents.map((agent) =>
agent.agentId === status.agentId
? {
...agent,
status: status.status,
currentTool: status.currentTool,
}
: agent,
),
}
},
)
} catch {}
})
return () => {
eventSource.close()
}
}, [ready, baseUrl, queryClient])
return query
}

View File

@@ -0,0 +1,71 @@
import { useInfiniteQuery } from '@tanstack/react-query'
import { useAgentServerUrl } from '@/lib/browseros/useBrowserOSProviders'
import type { AgentHistoryPageResponse } from './claw-chat-types'
const HISTORY_QUERY_KEY = 'claw-agent-history'
async function fetchClawJson<T>(url: string): Promise<T> {
const response = await fetch(url)
if (!response.ok) {
let message = `Request failed with status ${response.status}`
try {
const body = (await response.json()) as { error?: string }
if (body.error) message = body.error
} catch {}
throw new Error(message)
}
return response.json() as Promise<T>
}
function buildClawUrl(baseUrl: string, path: string): URL {
return new URL(`/claw${path}`, baseUrl)
}
export function useClawChatHistory({
agentId,
sessionKey,
enabled = true,
limit = 50,
}: {
agentId: string
// null lets the server resolve the most recent user-chat session for the
// agent — avoids an extra /session round-trip and the race that came with it.
sessionKey: string | null
enabled?: boolean
limit?: number
}) {
const {
baseUrl,
isLoading: urlLoading,
error: urlError,
} = useAgentServerUrl()
const query = useInfiniteQuery<AgentHistoryPageResponse, Error>({
queryKey: [HISTORY_QUERY_KEY, baseUrl, agentId, sessionKey],
initialPageParam: undefined as string | undefined,
queryFn: async ({ pageParam }) => {
const url = buildClawUrl(baseUrl as string, `/agents/${agentId}/history`)
url.searchParams.set('limit', String(limit))
if (sessionKey) {
url.searchParams.set('sessionKey', sessionKey)
}
if (typeof pageParam === 'string' && pageParam) {
url.searchParams.set('cursor', pageParam)
}
return fetchClawJson<AgentHistoryPageResponse>(url.toString())
},
getNextPageParam: (lastPage) =>
lastPage.page.hasMore ? lastPage.page.cursor : undefined,
enabled: enabled && Boolean(baseUrl) && !urlLoading && Boolean(agentId),
})
return {
...query,
error: query.error ?? urlError,
isLoading: query.isLoading || urlLoading,
}
}

View File

@@ -0,0 +1,270 @@
import { useCallback, useEffect, useRef, useState } from 'react'
import type { OpenClawChatHistoryMessage } from '@/entrypoints/app/agents/useOpenClaw'
import type { UserAttachmentPreview } from '@/lib/agent-conversations/types'
import type { ServerAttachmentPayload } from '@/lib/attachments'
import { useAgentServerUrl } from '@/lib/browseros/useBrowserOSProviders'
export type OutboundMessageStatus = 'queued' | 'sending' | 'failed'
export interface OutboundMessage {
id: string
text: string
attachments: ServerAttachmentPayload[]
attachmentPreviews: UserAttachmentPreview[]
status: OutboundMessageStatus
error?: string
createdAt: number
}
export interface OutboundQueueEnqueueInput {
text: string
attachments?: ServerAttachmentPayload[]
attachmentPreviews?: UserAttachmentPreview[]
history?: OpenClawChatHistoryMessage[]
}
export interface OutboundQueueApi {
queue: OutboundMessage[]
enqueue(input: OutboundQueueEnqueueInput): void
cancel(id: string): void
retry(id: string): void
}
interface UseOutboundQueueOptions {
agentId: string | null | undefined
sessionKey?: string | null
}
interface ServerQueuedItem {
id: string
status: 'queued' | 'dispatching' | 'failed'
message: string
attachmentsPreview: Array<{
kind: 'image' | 'file'
mediaType: string
name?: string
}>
error?: string
createdAt: number
}
function makeId(): string {
if (typeof crypto !== 'undefined' && crypto.randomUUID) {
return crypto.randomUUID()
}
return `${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 10)}`
}
/**
* Server-backed outbound message queue. The browser is purely a
* projection of server state — closing the tab is safe because the queue
* keeps draining server-side via the OutboundQueueService.
*
* Single id-keyed list: the client generates the queue id and hands it
* to the server in the POST body, so the optimistic row and the SSE
* snapshot reconcile on the same key from frame zero — there is no
* window in which the message renders twice.
*/
export function useOutboundQueue(
options: UseOutboundQueueOptions,
): OutboundQueueApi {
const { agentId, sessionKey } = options
const { baseUrl } = useAgentServerUrl()
const sessionKeyRef = useRef<string | null | undefined>(sessionKey)
sessionKeyRef.current = sessionKey
const [items, setItems] = useState<OutboundMessage[]>([])
// Track which ids the server has confirmed seeing in any SSE snapshot.
// We use this to know whether a missing-from-snapshot id is "drained
// by the server" (drop it) or "still in flight client-side" (keep
// showing the optimistic row).
const everSeenByServerRef = useRef<Set<string>>(new Set())
// Local-only attachment previews, keyed by queue id. Data URLs never
// leave the browser — the SSE feed only carries metadata, so we hold
// them here so the chip strip keeps rendering after server takeover.
const previewMapRef = useRef<Map<string, UserAttachmentPreview[]>>(new Map())
useEffect(() => {
if (!baseUrl || !agentId) {
setItems([])
everSeenByServerRef.current = new Set()
previewMapRef.current = new Map()
return
}
let cancelled = false
const url = `${baseUrl}/claw/agents/${encodeURIComponent(agentId)}/queue/stream`
const source = new EventSource(url)
source.onmessage = (event) => {
if (cancelled) return
try {
const parsed = JSON.parse(event.data) as { items: ServerQueuedItem[] }
const snapshotIds = new Set(parsed.items.map((item) => item.id))
for (const id of snapshotIds) everSeenByServerRef.current.add(id)
setItems((prev) => {
const next: OutboundMessage[] = parsed.items.map((item) => ({
id: item.id,
text: item.message,
attachments: [],
attachmentPreviews: previewMapRef.current.get(item.id) ?? [],
status: serverStatusToClient(item.status),
error: item.error,
createdAt: item.createdAt,
}))
// Carry forward any optimistic / failed entries the server
// doesn't know about yet (POST in flight) or has finished
// dispatching but the client wants to keep visible (failed).
const carried = prev.filter((local) => {
if (snapshotIds.has(local.id)) return false
if (everSeenByServerRef.current.has(local.id)) {
// Server saw it before and it's gone now — drained.
previewMapRef.current.delete(local.id)
return false
}
return local.status !== 'failed' || Boolean(local.error)
})
return [...carried, ...next]
})
} catch {
// Malformed event — ignore; next snapshot will recover.
}
}
source.onerror = () => {
// Auto-reconnects; nothing to do here.
}
return () => {
cancelled = true
source.close()
}
}, [baseUrl, agentId])
const enqueue = useCallback(
(input: OutboundQueueEnqueueInput) => {
if (!baseUrl || !agentId) return
const trimmed = input.text.trim()
const attachments = input.attachments ?? []
if (!trimmed && attachments.length === 0) return
const id = makeId()
const previews = input.attachmentPreviews ?? []
previewMapRef.current.set(id, previews)
setItems((prev) => [
...prev,
{
id,
text: trimmed,
attachments,
attachmentPreviews: previews,
status: 'queued',
createdAt: Date.now(),
},
])
void (async () => {
try {
const response = await fetch(
`${baseUrl}/claw/agents/${encodeURIComponent(agentId)}/queue`,
{
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({
id,
message: trimmed,
attachments: attachments.length > 0 ? attachments : undefined,
sessionKey: sessionKeyRef.current ?? undefined,
history: input.history,
}),
},
)
if (!response.ok) {
const text = await response.text().catch(() => '')
previewMapRef.current.delete(id)
setItems((prev) =>
prev.map((item) =>
item.id === id
? {
...item,
status: 'failed',
error:
text || `Failed to enqueue (status ${response.status})`,
}
: item,
),
)
}
} catch (err) {
// Only mark as failed if the SSE snapshot hasn't already
// taken ownership of the entry (i.e. the request actually
// reached the server).
if (everSeenByServerRef.current.has(id)) return
previewMapRef.current.delete(id)
setItems((prev) =>
prev.map((item) =>
item.id === id
? {
...item,
status: 'failed',
error:
err instanceof Error
? err.message
: 'Failed to enqueue message',
}
: item,
),
)
}
})()
},
[baseUrl, agentId],
)
const cancel = useCallback(
(id: string) => {
// If the server has never seen this id, just drop it locally.
if (!everSeenByServerRef.current.has(id)) {
previewMapRef.current.delete(id)
setItems((prev) => prev.filter((item) => item.id !== id))
return
}
if (!baseUrl || !agentId) return
void fetch(
`${baseUrl}/claw/agents/${encodeURIComponent(agentId)}/queue/${encodeURIComponent(id)}`,
{ method: 'DELETE' },
).catch(() => {})
},
[baseUrl, agentId],
)
const retry = useCallback(
(id: string) => {
if (!everSeenByServerRef.current.has(id)) {
// Optimistic-only entry, never made it to the server. Reset
// status so the user can press Send again.
setItems((prev) =>
prev.map((item) =>
item.id === id
? { ...item, status: 'queued', error: undefined }
: item,
),
)
return
}
if (!baseUrl || !agentId) return
void fetch(
`${baseUrl}/claw/agents/${encodeURIComponent(agentId)}/queue/${encodeURIComponent(id)}/retry`,
{ method: 'POST' },
).catch(() => {})
},
[baseUrl, agentId],
)
return { queue: items, enqueue, cancel, retry }
}
function serverStatusToClient(
status: ServerQueuedItem['status'],
): OutboundMessageStatus {
if (status === 'dispatching') return 'sending'
if (status === 'failed') return 'failed'
return 'queued'
}

View File

@@ -1,393 +0,0 @@
import {
ArrowLeft,
Bot,
CheckCircle2,
Loader2,
Send,
XCircle,
} from 'lucide-react'
import { type FC, useEffect, useRef, useState } from 'react'
import {
Message,
MessageContent,
MessageResponse,
} from '@/components/ai-elements/message'
import {
Reasoning,
ReasoningContent,
ReasoningTrigger,
} from '@/components/ai-elements/reasoning'
import { Button } from '@/components/ui/button'
import { Textarea } from '@/components/ui/textarea'
import { consumeSSEStream } from '@/lib/sse'
import { chatWithAgent, type OpenClawStreamEvent } from './useOpenClaw'
interface ToolEntry {
id: string
name: string
status: 'running' | 'completed' | 'error'
durationMs?: number
}
type AssistantPart =
| { kind: 'thinking'; text: string; done: boolean }
| { kind: 'tool-batch'; tools: ToolEntry[] }
| { kind: 'text'; text: string }
interface ChatTurn {
id: string
userText: string
parts: AssistantPart[]
done: boolean
}
interface AgentChatProps {
agentId: string
agentName: string
onBack: () => void
}
export const AgentChat: FC<AgentChatProps> = ({
agentId,
agentName,
onBack,
}) => {
const [turns, setTurns] = useState<ChatTurn[]>([])
const [input, setInput] = useState('')
const [streaming, setStreaming] = useState(false)
const scrollRef = useRef<HTMLDivElement>(null)
const sessionKeyRef = useRef(crypto.randomUUID())
const streamAbortRef = useRef<AbortController | null>(null)
const textAccRef = useRef('')
const thinkAccRef = useRef('')
const scrollToBottom = () => {
scrollRef.current?.scrollTo(0, scrollRef.current.scrollHeight)
}
// biome-ignore lint/correctness/useExhaustiveDependencies: scroll on every turns change
useEffect(() => {
scrollToBottom()
}, [turns])
useEffect(() => {
return () => {
streamAbortRef.current?.abort()
}
}, [])
const updateCurrentTurnParts = (
updater: (parts: AssistantPart[]) => AssistantPart[],
) => {
setTurns((prev) => {
const last = prev[prev.length - 1]
if (!last) return prev
return [...prev.slice(0, -1), { ...last, parts: updater(last.parts) }]
})
}
const processStreamEvent = (event: OpenClawStreamEvent) => {
switch (event.type) {
case 'text-delta': {
const delta = (event.data.text as string) ?? ''
textAccRef.current += delta
const text = textAccRef.current
updateCurrentTurnParts((parts) => {
const last = parts[parts.length - 1]
if (last?.kind === 'text') {
return [...parts.slice(0, -1), { ...last, text }]
}
return [...parts, { kind: 'text', text }]
})
break
}
case 'thinking': {
const delta = (event.data.text as string) ?? ''
thinkAccRef.current += delta
const text = thinkAccRef.current
updateCurrentTurnParts((parts) => {
const idx = parts.findIndex((p) => p.kind === 'thinking' && !p.done)
if (idx >= 0) {
return [
...parts.slice(0, idx),
{ ...parts[idx], text, done: false },
...parts.slice(idx + 1),
]
}
return [...parts, { kind: 'thinking', text, done: false }]
})
break
}
case 'tool-start': {
const tool: ToolEntry = {
id: (event.data.toolCallId as string) ?? crypto.randomUUID(),
name: (event.data.toolName as string) ?? 'unknown',
status: 'running',
}
updateCurrentTurnParts((parts) => {
const last = parts[parts.length - 1]
if (last?.kind === 'tool-batch') {
return [
...parts.slice(0, -1),
{ ...last, tools: [...last.tools, tool] },
]
}
return [...parts, { kind: 'tool-batch', tools: [tool] }]
})
break
}
case 'tool-end': {
const toolId = event.data.toolCallId as string
const status =
(event.data.status as string) === 'error' ? 'error' : 'completed'
const durationMs = event.data.durationMs as number | undefined
updateCurrentTurnParts((parts) => {
for (let i = parts.length - 1; i >= 0; i--) {
const part = parts[i]
if (
part.kind === 'tool-batch' &&
part.tools.some((t) => t.id === toolId)
) {
const updatedTools = part.tools.map((t) =>
t.id === toolId
? {
...t,
status: status as ToolEntry['status'],
durationMs,
}
: t,
)
return [
...parts.slice(0, i),
{ ...part, tools: updatedTools },
...parts.slice(i + 1),
]
}
}
return parts
})
break
}
case 'done': {
updateCurrentTurnParts((parts) =>
parts.map((part) =>
part.kind === 'thinking' ? { ...part, done: true } : part,
),
)
setTurns((prev) => {
const last = prev[prev.length - 1]
if (!last) return prev
return [...prev.slice(0, -1), { ...last, done: true }]
})
break
}
case 'error': {
const msg =
(event.data.message as string) ??
(event.data.error as string) ??
'Unknown error'
updateCurrentTurnParts((parts) => [
...parts,
{ kind: 'text', text: `Error: ${msg}` },
])
break
}
}
}
const handleSend = async () => {
const text = input.trim()
if (!text || streaming) return
const turn: ChatTurn = {
id: crypto.randomUUID(),
userText: text,
parts: [],
done: false,
}
setTurns((prev) => [...prev, turn])
setInput('')
setStreaming(true)
textAccRef.current = ''
thinkAccRef.current = ''
const abortController = new AbortController()
streamAbortRef.current = abortController
try {
const response = await chatWithAgent(
agentId,
text,
sessionKeyRef.current,
abortController.signal,
)
if (!response.ok) {
const err = await response.text()
updateCurrentTurnParts((parts) => [
...parts,
{ kind: 'text', text: `Error: ${err}` },
])
return
}
await consumeSSEStream(
response,
processStreamEvent,
abortController.signal,
)
} catch (err) {
if (abortController.signal.aborted) return
const msg = err instanceof Error ? err.message : String(err)
updateCurrentTurnParts((parts) => [
...parts,
{ kind: 'text', text: `Error: ${msg}` },
])
} finally {
if (streamAbortRef.current === abortController) {
streamAbortRef.current = null
}
setStreaming(false)
}
}
return (
<div className="flex h-[calc(100vh-4rem)] flex-col">
<div className="flex items-center gap-2 border-b px-4 py-3">
<Button variant="ghost" size="icon" onClick={onBack}>
<ArrowLeft className="size-4" />
</Button>
<h2 className="font-semibold text-lg">{agentName}</h2>
</div>
<div ref={scrollRef} className="flex-1 space-y-4 overflow-y-auto p-4">
{turns.map((turn) => (
<div key={turn.id} className="space-y-3">
{/* User message */}
<Message from="user">
<MessageContent>
<pre className="whitespace-pre-wrap font-sans text-sm">
{turn.userText}
</pre>
</MessageContent>
</Message>
{/* Assistant response — all parts grouped */}
{turn.parts.length > 0 && (
<Message from="assistant">
<MessageContent>
{turn.parts.map((part, i) => {
const key = `${turn.id}-part-${i}`
switch (part.kind) {
case 'thinking':
return (
<Reasoning
key={key}
className="w-full"
isStreaming={!part.done}
defaultOpen={!part.done}
>
<ReasoningTrigger />
<ReasoningContent>{part.text}</ReasoningContent>
</Reasoning>
)
case 'tool-batch':
return (
<div key={key} className="w-full space-y-1">
{part.tools.map((tool) => (
<div
key={tool.id}
className="flex items-center gap-2 rounded-md border px-3 py-2 text-sm"
>
{tool.status === 'running' && (
<Loader2 className="size-3.5 animate-spin text-muted-foreground" />
)}
{tool.status === 'completed' && (
<CheckCircle2 className="size-3.5 text-green-500" />
)}
{tool.status === 'error' && (
<XCircle className="size-3.5 text-destructive" />
)}
<span className="font-mono text-xs">
{tool.name}
</span>
{tool.durationMs != null && (
<span className="ml-auto text-muted-foreground text-xs">
{(tool.durationMs / 1000).toFixed(1)}s
</span>
)}
</div>
))}
</div>
)
case 'text':
return (
<MessageResponse key={key}>
{part.text}
</MessageResponse>
)
default:
return null
}
})}
</MessageContent>
</Message>
)}
{/* Streaming indicator when waiting for first part */}
{!turn.done && turn.parts.length === 0 && streaming && (
<div className="flex gap-2">
<div className="flex h-7 w-7 shrink-0 items-center justify-center rounded-full bg-[var(--accent-orange)] text-white">
<Bot className="h-3.5 w-3.5" />
</div>
<div className="flex items-center gap-1 rounded-xl rounded-tl-none border border-border/50 bg-card px-3 py-2.5 shadow-sm">
<span className="h-1.5 w-1.5 animate-bounce rounded-full bg-[var(--accent-orange)] [animation-delay:-0.3s]" />
<span className="h-1.5 w-1.5 animate-bounce rounded-full bg-[var(--accent-orange)] [animation-delay:-0.15s]" />
<span className="h-1.5 w-1.5 animate-bounce rounded-full bg-[var(--accent-orange)]" />
</div>
</div>
)}
</div>
))}
</div>
<div className="border-t p-4">
<div className="flex gap-2">
<Textarea
value={input}
onChange={(e) => setInput(e.target.value)}
onKeyDown={(e) => {
if (e.key === 'Enter' && !e.shiftKey) {
e.preventDefault()
handleSend()
}
}}
placeholder="Send a message..."
className="min-h-[44px] resize-none"
rows={1}
/>
<Button
onClick={handleSend}
disabled={!input.trim() || streaming}
size="icon"
>
{streaming ? (
<Loader2 className="size-4 animate-spin" />
) : (
<Send className="size-4" />
)}
</Button>
</div>
</div>
</div>
)
}

View File

@@ -5,14 +5,16 @@ import {
import { FitAddon } from '@xterm/addon-fit'
import { WebLinksAddon } from '@xterm/addon-web-links'
import { Terminal } from '@xterm/xterm'
import { ArrowLeft } from 'lucide-react'
import { type FC, useEffect, useRef } from 'react'
import { ArrowLeft, Check, Copy } from 'lucide-react'
import { type FC, useEffect, useRef, useState } from 'react'
import '@xterm/xterm/css/xterm.css'
import { Button } from '@/components/ui/button'
import { getAgentServerUrl } from '@/lib/browseros/helpers'
interface AgentTerminalProps {
onBack: () => void
initialCommand?: string
onSessionExit?: () => void
}
type TerminalServerMessage =
@@ -36,26 +38,22 @@ function resolveCssColor(variableName: string): string {
return color
}
function withAlpha(color: string, alpha: number): string {
const channels = color.match(/[\d.]+/g)
if (!channels || channels.length < 3) return color
const [red, green, blue] = channels
return `rgb(${red} ${green} ${blue} / ${alpha})`
}
function createTerminalTheme() {
const isDark = document.documentElement.classList.contains('dark')
const background = resolveCssColor('--background')
const foreground = resolveCssColor('--foreground')
const muted = resolveCssColor('--muted-foreground')
const accent = resolveCssColor('--accent-orange')
return {
background,
foreground,
cursor: foreground,
cursorAccent: background,
selectionBackground: withAlpha(accent, isDark ? 0.3 : 0.2),
// Solid terminal-standard selection colors. Deriving from a CSS var
// with alpha composed against the background produced near-white
// rectangles on light mode, making selection invisible.
selectionBackground: isDark ? '#3a4463' : '#b4d4f4',
selectionInactiveBackground: isDark ? '#2b3348' : '#d9e5f3',
selectionForeground: foreground,
black: isDark ? '#16131a' : '#1f1b22',
red: isDark ? '#ef8c7c' : '#c25544',
@@ -118,8 +116,38 @@ function parseTerminalMessage(data: unknown): TerminalServerMessage | null {
return null
}
export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
export const AgentTerminal: FC<AgentTerminalProps> = ({
onBack,
initialCommand,
onSessionExit,
}) => {
const containerRef = useRef<HTMLDivElement>(null)
const terminalRef = useRef<Terminal | null>(null)
// Refs keep the mount-once effect from tearing down the PTY when the
// parent re-renders with new inline callbacks.
const initialCommandRef = useRef(initialCommand)
const onSessionExitRef = useRef(onSessionExit)
initialCommandRef.current = initialCommand
onSessionExitRef.current = onSessionExit
const [copied, setCopied] = useState(false)
// Copy the current xterm selection to the browser clipboard. No-op
// if nothing is selected — users who want the whole buffer can
// Cmd+A first. Uses the browser clipboard, not the container's, so
// it works even when the running TUI has mouse tracking enabled
// (Opt+drag forces a selection regardless, see terminal config).
const handleCopy = async (): Promise<void> => {
const text = terminalRef.current?.getSelection()
if (!text) return
try {
await navigator.clipboard.writeText(text)
setCopied(true)
window.setTimeout(() => setCopied(false), 1500)
} catch {
// clipboard permission denied or unavailable — swallow, user will retry
}
}
useEffect(() => {
if (!containerRef.current) return
@@ -132,6 +160,34 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
lineHeight: 1.25,
scrollback: 8000,
theme: createTerminalTheme(),
// Opt+click+drag forces a native text selection even when the
// running TUI has mouse-tracking enabled (xterm would otherwise
// forward every click to the app and selection wouldn't work).
macOptionClickForcesSelection: true,
})
terminalRef.current = terminal
// Cmd+A → select all, Cmd+C → copy selection via the browser
// clipboard. Return false so xterm doesn't also forward the keys
// to the running program.
terminal.attachCustomKeyEventHandler((event) => {
if (event.type !== 'keydown') return true
const isMac = navigator.platform.toUpperCase().includes('MAC')
const mod = isMac ? event.metaKey : event.ctrlKey
if (!mod) return true
const key = event.key.toLowerCase()
if (key === 'a') {
terminal.selectAll()
return false
}
if (key === 'c') {
const sel = terminal.getSelection()
if (sel) {
void navigator.clipboard.writeText(sel)
return false
}
}
return true
})
const fitAddon = new FitAddon()
@@ -139,6 +195,12 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
terminal.loadAddon(new WebLinksAddon())
terminal.open(containerRef.current)
// React 18 StrictMode double-invokes effects in dev. Everything
// async inside this effect is scoped to an AbortController; the
// cleanup aborts it and any pending awaits bail out, so we never
// leak a second live WebSocket or duplicate xterm listeners.
const ac = new AbortController()
const cleanups: Array<() => void> = []
let ws: WebSocket | null = null
let sawExit = false
@@ -159,17 +221,28 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
sendMessage({ type: 'resize', cols, rows })
}
const connect = async () => {
const connect = async (): Promise<void> => {
const baseUrl = await getAgentServerUrl()
if (ac.signal.aborted) return
const wsUrl = new URL('/terminal/ws', baseUrl)
wsUrl.protocol = wsUrl.protocol === 'https:' ? 'wss:' : 'ws:'
ws = new WebSocket(wsUrl)
// If the effect was cleaned up between the await above and now,
// close the socket we just opened and bail.
if (ac.signal.aborted) {
ws.close()
ws = null
return
}
cleanups.push(() => ws?.close())
ws.onopen = () => {
fitAddon.fit()
terminal.focus()
sendResize()
const cmd = initialCommandRef.current
if (cmd) sendMessage({ type: 'input', data: `${cmd}\n` })
}
ws.onmessage = (event) => {
@@ -185,6 +258,7 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
terminal.write(
`\r\n\x1b[90m[session ended with exit ${message.exitCode}]\x1b[0m\r\n`,
)
onSessionExitRef.current?.()
}
}
@@ -200,49 +274,41 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
const inputDisposable = terminal.onData((data) => {
sendMessage({ type: 'input', data })
})
const resizeDisposable = terminal.onResize(({ cols, rows }) => {
sendResize(cols, rows)
})
return () => {
inputDisposable.dispose()
resizeDisposable.dispose()
}
cleanups.push(() => inputDisposable.dispose())
cleanups.push(() => resizeDisposable.dispose())
}
let disposeSocketBindings: (() => void) | undefined
void connect().then((disposeBindings) => {
disposeSocketBindings = disposeBindings
})
void connect()
const resizeObserver = new ResizeObserver(() => {
fitAddon.fit()
sendResize()
})
resizeObserver.observe(containerRef.current)
cleanups.push(() => resizeObserver.disconnect())
const themeObserver = new MutationObserver(() => {
applyTheme()
})
const themeObserver = new MutationObserver(() => applyTheme())
themeObserver.observe(document.documentElement, {
attributes: true,
attributeFilter: ['class'],
})
cleanups.push(() => themeObserver.disconnect())
return () => {
resizeObserver.disconnect()
themeObserver.disconnect()
disposeSocketBindings?.()
ws?.close()
ac.abort()
for (const dispose of cleanups) dispose()
terminal.dispose()
terminalRef.current = null
}
}, [])
return (
<div className="flex h-[calc(100dvh-10rem)] min-h-[32rem] w-full flex-col py-2 sm:min-h-[42rem] sm:py-4">
<div className="flex min-h-0 flex-1 flex-col overflow-hidden rounded-xl border border-border bg-card shadow-sm">
<div className="flex items-center gap-3 border-border border-b px-4 py-3 sm:px-6">
<div className="flex items-center justify-between gap-3 border-border border-b px-4 py-3 sm:px-6">
<div className="flex min-w-0 items-center gap-3">
<Button variant="ghost" size="icon" onClick={onBack}>
<ArrowLeft className="size-4" />
@@ -256,6 +322,14 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
</div>
</div>
</div>
<Button variant="outline" size="sm" onClick={handleCopy}>
{copied ? (
<Check className="mr-1 size-3.5" />
) : (
<Copy className="mr-1 size-3.5" />
)}
{copied ? 'Copied' : 'Copy'}
</Button>
</div>
<div className="min-h-0 flex-1 p-4 sm:p-6">
@@ -269,7 +343,7 @@ export const AgentTerminal: FC<AgentTerminalProps> = ({ onBack }) => {
</div>
</div>
<div className="min-h-0 flex-1 px-4 py-4 sm:px-5 sm:py-5">
<div className="min-h-0 flex-1 cursor-text px-4 py-4 sm:px-5 sm:py-5">
<div ref={containerRef} className="h-full w-full" />
</div>
</div>

View File

@@ -0,0 +1,185 @@
import { useQuery } from '@tanstack/react-query'
import { CheckCircle2, Loader2, Terminal, TriangleAlert } from 'lucide-react'
import type { FC } from 'react'
import { Button } from '@/components/ui/button'
import { useAgentServerUrl } from '@/lib/browseros/useBrowserOSProviders'
export interface OpenClawCliProvider {
id: string
displayName: string
description: string
models: readonly string[]
authLoginCommand: string
}
export interface OpenClawCliProviderAuthStatus {
installed: boolean
loggedIn: boolean
accountLabel?: string
subscriptionLabel?: string
error?: string
}
export interface OpenClawCliProviderOption {
id: string
type: string
name: string
modelId: string
}
const CLAUDE_CLI_PROVIDER: OpenClawCliProvider = {
id: 'claude-cli',
displayName: 'Anthropic Claude CLI',
description: 'Uses your Claude.ai subscription via the Claude Code CLI',
models: ['claude-sonnet-4-6', 'claude-opus-4-6', 'claude-haiku-4-5'],
authLoginCommand: 'claude /login',
}
export const OPENCLAW_CLI_PROVIDERS: readonly OpenClawCliProvider[] = [
CLAUDE_CLI_PROVIDER,
]
export function findOpenClawCliProviderById(
id: string,
): OpenClawCliProvider | undefined {
return OPENCLAW_CLI_PROVIDERS.find((provider) => provider.id === id)
}
export function buildOpenClawCliProviderOptions(): OpenClawCliProviderOption[] {
return OPENCLAW_CLI_PROVIDERS.flatMap((provider) =>
provider.models.map((modelId) => ({
id: `${provider.id}/${modelId}`,
type: provider.id,
name: provider.displayName,
modelId,
})),
)
}
async function fetchCliProviderAuthStatus(
baseUrl: string,
providerId: string,
): Promise<OpenClawCliProviderAuthStatus> {
const res = await fetch(`${baseUrl}/claw/providers/${providerId}/auth-status`)
if (!res.ok) {
let message = `Auth status request failed (${res.status})`
try {
const body = (await res.json()) as { error?: string }
if (body.error) message = body.error
} catch {}
throw new Error(message)
}
return res.json() as Promise<OpenClawCliProviderAuthStatus>
}
export function useOpenClawCliProviderAuthStatus(
providerId: string,
enabled: boolean,
) {
const { baseUrl, isLoading: urlLoading } = useAgentServerUrl()
return useQuery<OpenClawCliProviderAuthStatus, Error>({
queryKey: ['openclaw-cli-auth', baseUrl, providerId],
queryFn: () => fetchCliProviderAuthStatus(baseUrl as string, providerId),
enabled: !!baseUrl && !urlLoading && enabled,
refetchInterval: enabled ? 2000 : false,
})
}
interface OpenClawCliProviderStatusPanelProps {
provider: OpenClawCliProvider
status: OpenClawCliProviderAuthStatus | undefined
loading: boolean
fetchError: Error | null
onConnect: () => void
}
export const OpenClawCliProviderStatusPanel: FC<
OpenClawCliProviderStatusPanelProps
> = ({ provider, status, loading, fetchError, onConnect }) => {
// Initial fetch (no data yet).
if (loading && !status) {
return (
<div className="flex items-center gap-2 rounded-md border border-border bg-muted/30 px-3 py-2 text-sm">
<Loader2 className="size-4 animate-spin text-muted-foreground" />
<span className="text-muted-foreground">
Checking {provider.displayName} status
</span>
</div>
)
}
if (fetchError) {
return (
<div className="flex items-start gap-2 rounded-md border border-destructive/30 bg-destructive/5 px-3 py-2 text-sm">
<TriangleAlert className="mt-0.5 size-4 text-destructive" />
<div>
<div className="font-medium text-destructive">
Could not read {provider.displayName} status
</div>
<div className="text-muted-foreground text-xs">
{fetchError.message}
</div>
</div>
</div>
)
}
if (!status) return null
// Install failed or binary missing.
if (!status.installed) {
return (
<div className="flex items-start gap-2 rounded-md border border-amber-500/40 bg-amber-500/5 px-3 py-2 text-sm">
<TriangleAlert className="mt-0.5 size-4 text-amber-600" />
<div>
<div className="font-medium">
{provider.displayName} not installed
</div>
<div className="text-muted-foreground text-xs">
The gateway will try to install it on the next restart. If this
persists, check your network and the gateway logs.
</div>
</div>
</div>
)
}
// Happy path.
if (status.loggedIn) {
const identityBits = [
status.accountLabel,
status.subscriptionLabel ? `(${status.subscriptionLabel})` : null,
].filter(Boolean)
const identity = identityBits.length > 0 ? identityBits.join(' ') : 'Ready'
return (
<div className="flex items-center gap-2 rounded-md border border-emerald-500/40 bg-emerald-500/5 px-3 py-2 text-sm">
<CheckCircle2 className="size-4 text-emerald-600" />
<div className="min-w-0 flex-1">
<div className="font-medium">Connected to {provider.displayName}</div>
<div className="truncate text-muted-foreground text-xs">
{identity}
</div>
</div>
</div>
)
}
// Installed but not logged in.
return (
<div className="flex flex-col gap-2 rounded-md border border-border bg-muted/30 px-3 py-3 text-sm">
<div>
<div className="font-medium">{provider.displayName} not set up</div>
<div className="text-muted-foreground text-xs">
{provider.description}
</div>
{status.error && (
<div className="mt-1 text-destructive text-xs">{status.error}</div>
)}
</div>
<Button size="sm" variant="outline" onClick={onConnect} className="w-fit">
<Terminal className="mr-1 size-4" />
Connect {provider.displayName}
</Button>
</div>
)
}

View File

@@ -0,0 +1,24 @@
import type { LlmProviderConfig, ProviderType } from '@/lib/llm-providers/types'
const OPENCLAW_SUPPORTED_PROVIDER_TYPES: ProviderType[] = [
'openrouter',
'openai',
'openai-compatible',
'anthropic',
'moonshot',
]
export function isOpenClawSupportedProviderType(
providerType: ProviderType,
): boolean {
return OPENCLAW_SUPPORTED_PROVIDER_TYPES.includes(providerType)
}
export function getOpenClawSupportedProviders(
providers: LlmProviderConfig[],
): LlmProviderConfig[] {
return providers.filter(
(provider) =>
!!provider.apiKey && isOpenClawSupportedProviderType(provider.type),
)
}

View File

@@ -1,7 +1,3 @@
import type {
BrowserOSAgentRoleId,
BrowserOSCustomRoleInput,
} from '@browseros/shared/types/role-aware-agents'
import { useMutation, useQuery, useQueryClient } from '@tanstack/react-query'
import { getAgentServerUrl } from '@/lib/browseros/helpers'
import { useAgentServerUrl } from '@/lib/browseros/useBrowserOSProviders'
@@ -11,27 +7,6 @@ export interface AgentEntry {
name: string
workspace: string
model?: unknown
role?: {
roleSource: 'builtin' | 'custom'
roleId?: BrowserOSAgentRoleId
roleName: string
shortDescription: string
}
}
export interface RoleTemplateSummary {
id: BrowserOSAgentRoleId
name: string
shortDescription: string
longDescription: string
recommendedApps: string[]
defaultAgentName: string
boundaries: Array<{
key: string
label: string
description: string
defaultMode: 'allow' | 'ask' | 'block'
}>
}
export interface OpenClawStatus {
@@ -61,8 +36,6 @@ export interface OpenClawStatus {
export interface OpenClawAgentMutationInput {
name: string
roleId?: BrowserOSAgentRoleId
customRole?: BrowserOSCustomRoleInput
providerType?: string
providerName?: string
baseUrl?: string
@@ -86,9 +59,15 @@ export function getModelDisplayName(model: unknown): string | undefined {
export const OPENCLAW_QUERY_KEYS = {
status: 'openclaw-status',
agents: 'openclaw-agents',
roles: 'openclaw-roles',
} as const
export type GatewayLifecycleAction =
| 'setup'
| 'start'
| 'stop'
| 'restart'
| 'reconnect'
async function clawFetch<T>(
baseUrl: string,
path: string,
@@ -117,16 +96,6 @@ async function fetchOpenClawAgents(baseUrl: string): Promise<AgentEntry[]> {
return data.agents ?? []
}
async function fetchOpenClawRoles(
baseUrl: string,
): Promise<RoleTemplateSummary[]> {
const data = await clawFetch<{ roles: RoleTemplateSummary[] }>(
baseUrl,
'/roles',
)
return data.roles ?? []
}
async function invalidateOpenClawQueries(
queryClient: ReturnType<typeof useQueryClient>,
): Promise<void> {
@@ -179,28 +148,6 @@ export function useOpenClawAgents(enabled = true) {
}
}
export function useOpenClawRoles() {
const {
baseUrl,
isLoading: urlLoading,
error: urlError,
} = useAgentServerUrl()
const query = useQuery<RoleTemplateSummary[], Error>({
queryKey: [OPENCLAW_QUERY_KEYS.roles, baseUrl],
queryFn: () => fetchOpenClawRoles(baseUrl as string),
enabled: !!baseUrl && !urlLoading,
staleTime: 60_000,
})
return {
roles: query.data ?? [],
loading: query.isLoading || urlLoading,
error: query.error ?? urlError,
refetch: query.refetch,
}
}
export function useOpenClawMutations() {
const { baseUrl, isLoading: urlLoading } = useAgentServerUrl()
const queryClient = useQueryClient()
@@ -278,6 +225,13 @@ export function useOpenClawMutations() {
onSuccess,
})
let pendingGatewayAction: GatewayLifecycleAction | null = null
if (setupMutation.isPending) pendingGatewayAction = 'setup'
else if (restartMutation.isPending) pendingGatewayAction = 'restart'
else if (stopMutation.isPending) pendingGatewayAction = 'stop'
else if (startMutation.isPending) pendingGatewayAction = 'start'
else if (reconnectMutation.isPending) pendingGatewayAction = 'reconnect'
return {
setupOpenClaw: setupMutation.mutateAsync,
createAgent: createMutation.mutateAsync,
@@ -298,6 +252,7 @@ export function useOpenClawMutations() {
creating: createMutation.isPending,
deleting: deleteMutation.isPending,
reconnecting: reconnectMutation.isPending,
pendingGatewayAction,
}
}
@@ -314,17 +269,66 @@ export interface OpenClawStreamEvent {
data: Record<string, unknown>
}
export interface OpenClawChatHistoryMessage {
role: 'user' | 'assistant'
content: string
}
interface ChatHistoryTurnLike {
userText: string
parts: Array<{ kind: string; text?: string }>
}
export function buildChatHistoryFromTurns(
turns: ChatHistoryTurnLike[],
): OpenClawChatHistoryMessage[] {
const messages: OpenClawChatHistoryMessage[] = []
for (const turn of turns) {
const userText = turn.userText.trim()
if (userText) {
messages.push({ role: 'user', content: userText })
}
const assistantText = turn.parts
.filter(
(
part,
): part is {
kind: 'text'
text: string
} => part.kind === 'text' && typeof part.text === 'string',
)
.map((part) => part.text.trim())
.filter(Boolean)
.join('\n\n')
if (assistantText) {
messages.push({ role: 'assistant', content: assistantText })
}
}
return messages
}
export async function chatWithAgent(
agentId: string,
message: string,
sessionKey?: string,
history: OpenClawChatHistoryMessage[] = [],
signal?: AbortSignal,
attachments?: ReadonlyArray<unknown>,
): Promise<Response> {
const baseUrl = await getAgentServerUrl()
return fetch(`${baseUrl}/claw/agents/${agentId}/chat`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ message, sessionKey }),
body: JSON.stringify({
message,
sessionKey,
history,
...(attachments && attachments.length > 0 ? { attachments } : {}),
}),
signal,
})
}

View File

@@ -18,8 +18,8 @@ describe('route-utils', () => {
expect(shouldUseChatSession('/home/chat')).toBe(true)
})
it('keeps the focus grid on home while hiding it on dedicated full-screen routes', () => {
expect(shouldHideFocusGrid('/home')).toBe(false)
it('hides the focus grid on full-screen routes', () => {
expect(shouldHideFocusGrid('/home')).toBe(true)
expect(shouldHideFocusGrid('/home/agents/main')).toBe(true)
expect(shouldHideFocusGrid('/home/chat')).toBe(true)
expect(shouldHideFocusGrid('/home/skills')).toBe(true)

View File

@@ -1,4 +1,5 @@
const HIDE_FOCUS_GRID_PATHS = new Set([
'/home',
'/home/soul',
'/home/memory',
'/home/skills',

View File

@@ -12,6 +12,8 @@ export interface AssistantThinkingPart {
export interface ToolEntry {
id: string
name: string
label: string
subject?: string
status: 'running' | 'completed' | 'error'
durationMs?: number
}
@@ -26,9 +28,24 @@ export type AssistantPart =
| AssistantThinkingPart
| AssistantToolBatchPart
/**
* Attachments rendered alongside the user's text on the optimistic turn
* — populated when the composer staged any images/files. The dataUrl is
* the same one the server received; we keep it in memory only for the
* lifetime of the live turn (history reload re-fetches via the JSONL).
*/
export interface UserAttachmentPreview {
id: string
kind: 'image' | 'file'
mediaType: string
name: string
dataUrl?: string
}
export interface AgentConversationTurn {
id: string
userText: string
userAttachments?: UserAttachmentPreview[]
parts: AssistantPart[]
done: boolean
timestamp: number
@@ -50,4 +67,7 @@ export interface AgentCardData {
status: 'idle' | 'working' | 'error'
lastMessage?: string
lastMessageTimestamp?: number
activitySummary?: string
currentTool?: string
costUsd?: number
}

View File

@@ -0,0 +1,369 @@
/**
* Composer attachment helpers — validation, image compression, and the
* client-side payload shape sent to /agents/:id/chat.
*
* Image attachments travel as `data:` URLs (base64) so the gateway, which
* runs on 127.0.0.1 over Lima virtiofs, can ingest them as standard
* OpenAI-style content blocks. Non-image text-shaped files are read into
* memory and travel as their extracted text body — the server inlines
* them as a fenced `<attachment>` block on the user message.
*/
export const MAX_ATTACHMENTS_PER_MESSAGE = 10
export const MAX_IMAGE_BYTES = 5 * 1024 * 1024 // 5 MB after compression
export const MAX_FILE_TEXT_BYTES = 1 * 1024 * 1024 // 1 MB extracted text
export const IMAGE_LONG_EDGE_CAP = 2048
export const ALLOWED_IMAGE_MEDIA_TYPES = [
'image/png',
'image/jpeg',
'image/jpg',
'image/webp',
'image/gif',
] as const
export const ALLOWED_FILE_MEDIA_TYPE_PREFIXES = [
'text/',
'application/json',
] as const
export type ServerImageAttachment = {
kind: 'image'
mediaType: string
dataUrl: string
name?: string
}
export type ServerFileAttachment = {
kind: 'file'
mediaType: string
name: string
text: string
}
export type ServerAttachmentPayload =
| ServerImageAttachment
| ServerFileAttachment
/** UI-side representation: what the composer needs to render a chip. */
export interface StagedAttachment {
id: string
kind: 'image' | 'file'
mediaType: string
name: string
// Set for images so the chip thumbnail can render directly. For files
// we don't need a preview yet, but the field exists for v2 PDF previews.
dataUrl?: string
// Pre-computed payload for the server. Built once at staging time so
// re-renders don't re-encode large blobs.
payload: ServerAttachmentPayload
}
export type AttachmentValidationError =
| { code: 'too_many'; message: string }
| { code: 'unsupported_type'; message: string; mediaType: string }
| { code: 'too_large'; message: string }
| { code: 'read_failed'; message: string }
export type StageAttachmentResult =
| { ok: true; attachment: StagedAttachment }
| { ok: false; error: AttachmentValidationError }
function isImageMediaType(mediaType: string): boolean {
return (ALLOWED_IMAGE_MEDIA_TYPES as readonly string[]).includes(mediaType)
}
function isAllowedFileMediaType(mediaType: string): boolean {
return ALLOWED_FILE_MEDIA_TYPE_PREFIXES.some((prefix) =>
mediaType.startsWith(prefix),
)
}
/** Build a unique id without depending on `crypto.randomUUID` outside DOM. */
function makeId(): string {
if (typeof crypto !== 'undefined' && crypto.randomUUID) {
return crypto.randomUUID()
}
return `att-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 10)}`
}
/**
* Read a `File` and produce the staged-attachment shape — validate type,
* compress if it's a large image, and pre-build the server payload.
*/
export async function stageAttachment(
file: File,
): Promise<StageAttachmentResult> {
const mediaType = file.type || 'application/octet-stream'
if (isImageMediaType(mediaType)) {
try {
const compressed = await compressImageIfNeeded(file)
const dataUrl = await readAsDataUrl(compressed)
// Rough byte ceiling — `data:image/png;base64,...` doubles size with
// base64. Reject early so we never POST something the route will 400.
if (dataUrl.length > MAX_IMAGE_BYTES * 2) {
return {
ok: false,
error: {
code: 'too_large',
message: `Image "${file.name}" is too large (max ${humanBytes(
MAX_IMAGE_BYTES,
)}).`,
},
}
}
return {
ok: true,
attachment: {
id: makeId(),
kind: 'image',
mediaType,
name: file.name || 'image',
dataUrl,
payload: {
kind: 'image',
mediaType,
dataUrl,
name: file.name || undefined,
},
},
}
} catch (err) {
return {
ok: false,
error: {
code: 'read_failed',
message:
err instanceof Error
? err.message
: `Failed to read image "${file.name}".`,
},
}
}
}
if (isAllowedFileMediaType(mediaType)) {
let text: string
try {
text = await file.text()
} catch (err) {
return {
ok: false,
error: {
code: 'read_failed',
message:
err instanceof Error
? err.message
: `Failed to read file "${file.name}".`,
},
}
}
if (text.length > MAX_FILE_TEXT_BYTES) {
return {
ok: false,
error: {
code: 'too_large',
message: `File "${file.name}" is too large (max ${humanBytes(
MAX_FILE_TEXT_BYTES,
)}).`,
},
}
}
return {
ok: true,
attachment: {
id: makeId(),
kind: 'file',
mediaType,
name: file.name || 'attachment',
payload: {
kind: 'file',
mediaType,
name: file.name || 'attachment',
text,
},
},
}
}
return {
ok: false,
error: {
code: 'unsupported_type',
message: `Unsupported attachment type: ${mediaType || 'unknown'}`,
mediaType,
},
}
}
/**
* Stage multiple files at once, enforcing the per-message cap. The result
* partitions successful stages and any errors so the caller can show
* granular toasts.
*/
export async function stageAttachments(
files: File[],
alreadyStaged: number,
): Promise<{
staged: StagedAttachment[]
errors: AttachmentValidationError[]
}> {
const remainingSlots = Math.max(
0,
MAX_ATTACHMENTS_PER_MESSAGE - alreadyStaged,
)
const staged: StagedAttachment[] = []
const errors: AttachmentValidationError[] = []
if (remainingSlots === 0 && files.length > 0) {
errors.push({
code: 'too_many',
message: `At most ${MAX_ATTACHMENTS_PER_MESSAGE} attachments per message.`,
})
return { staged, errors }
}
const overflow = files.length - remainingSlots
if (overflow > 0) {
errors.push({
code: 'too_many',
message: `Only the first ${remainingSlots} of ${files.length} files were attached (max ${MAX_ATTACHMENTS_PER_MESSAGE}).`,
})
}
for (const file of files.slice(0, remainingSlots)) {
const result = await stageAttachment(file)
if (result.ok) {
staged.push(result.attachment)
} else {
errors.push(result.error)
}
}
return { staged, errors }
}
/**
* Resize images that are oversized to a sane long-edge cap. JPEG/WebP
* source files are re-encoded to JPEG; PNGs/GIFs that are already small
* are passed through untouched.
*/
export async function compressImageIfNeeded(file: File): Promise<Blob> {
// Cheap path: small files don't need any transform.
if (file.size <= 1.5 * 1024 * 1024) return file
const bitmap = await blobToImageBitmap(file)
const { width, height } = bitmap
const longEdge = Math.max(width, height)
if (longEdge <= IMAGE_LONG_EDGE_CAP && file.size <= MAX_IMAGE_BYTES) {
bitmap.close?.()
return file
}
const scale = Math.min(1, IMAGE_LONG_EDGE_CAP / longEdge)
const targetWidth = Math.max(1, Math.round(width * scale))
const targetHeight = Math.max(1, Math.round(height * scale))
const canvas =
typeof OffscreenCanvas !== 'undefined'
? new OffscreenCanvas(targetWidth, targetHeight)
: Object.assign(document.createElement('canvas'), {
width: targetWidth,
height: targetHeight,
})
const ctx = canvas.getContext('2d') as
| CanvasRenderingContext2D
| OffscreenCanvasRenderingContext2D
| null
if (!ctx) {
bitmap.close?.()
return file
}
ctx.drawImage(bitmap, 0, 0, targetWidth, targetHeight)
bitmap.close?.()
const outputType = 'image/jpeg'
if (canvas instanceof HTMLCanvasElement) {
return await new Promise<Blob>((resolve, reject) => {
canvas.toBlob(
(blob) => {
if (blob) resolve(blob)
else reject(new Error('Image compression failed.'))
},
outputType,
0.85,
)
})
}
return await (canvas as OffscreenCanvas).convertToBlob({
type: outputType,
quality: 0.85,
})
}
async function blobToImageBitmap(blob: Blob): Promise<ImageBitmap> {
if (typeof createImageBitmap === 'function') {
return createImageBitmap(blob)
}
// Fallback: load via an Image element and use the canvas decode path.
const url = URL.createObjectURL(blob)
try {
const img = await new Promise<HTMLImageElement>((resolve, reject) => {
const el = new Image()
el.onload = () => resolve(el)
el.onerror = () =>
reject(new Error('Failed to decode image for compression.'))
el.src = url
})
const canvas = document.createElement('canvas')
canvas.width = img.naturalWidth
canvas.height = img.naturalHeight
const ctx = canvas.getContext('2d')
if (!ctx) throw new Error('Canvas 2D context unavailable.')
ctx.drawImage(img, 0, 0)
const blobOut = await new Promise<Blob | null>((resolve) =>
canvas.toBlob(resolve, 'image/png'),
)
if (!blobOut) throw new Error('Canvas toBlob returned null.')
return await createImageBitmap(blobOut)
} finally {
URL.revokeObjectURL(url)
}
}
async function readAsDataUrl(blob: Blob): Promise<string> {
if ('arrayBuffer' in blob && typeof FileReader === 'undefined') {
const buffer = await blob.arrayBuffer()
const base64 = arrayBufferToBase64(buffer)
const type = blob.type || 'application/octet-stream'
return `data:${type};base64,${base64}`
}
return await new Promise<string>((resolve, reject) => {
const reader = new FileReader()
reader.onload = () => resolve(reader.result as string)
reader.onerror = () =>
reject(reader.error ?? new Error('FileReader failed to read blob.'))
reader.readAsDataURL(blob)
})
}
function arrayBufferToBase64(buffer: ArrayBuffer): string {
const bytes = new Uint8Array(buffer)
let binary = ''
const chunkSize = 0x8000
for (let i = 0; i < bytes.byteLength; i += chunkSize) {
binary += String.fromCharCode.apply(
null,
Array.from(bytes.subarray(i, Math.min(i + chunkSize, bytes.byteLength))),
)
}
return btoa(binary)
}
function humanBytes(bytes: number): string {
if (bytes >= 1024 * 1024) return `${(bytes / 1024 / 1024).toFixed(0)} MB`
if (bytes >= 1024) return `${(bytes / 1024).toFixed(0)} KB`
return `${bytes} B`
}

View File

@@ -0,0 +1,325 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Maps raw tool names + arguments to human-readable activity labels for
* the chat UI activity view. The MCP ToolRegistry is the source of truth
* for tool *existence*; this file is the editorial layer that turns
* snake_case identifiers into agent-speak verbs.
*/
const VERB_OVERRIDES: Record<string, string> = {
// Navigation
navigate_page: 'Navigated to',
new_page: 'Opened tab',
new_hidden_page: 'Opened tab',
show_page: 'Showed tab',
close_page: 'Closed tab',
list_pages: 'Listed open tabs',
get_active_page: 'Got active tab',
move_page: 'Moved tab',
group_tabs: 'Grouped tabs',
// Page reading
take_snapshot: 'Captured page snapshot',
take_enhanced_snapshot: 'Captured detailed snapshot',
get_page_content: 'Read page content',
get_page_links: 'Extracted page links',
get_dom: 'Read page DOM',
search_dom: 'Searched page DOM',
take_screenshot: 'Took screenshot',
// Input
click: 'Clicked',
click_at: 'Clicked at coordinates',
hover: 'Hovered',
hover_at: 'Hovered at coordinates',
type_at: 'Typed at coordinates',
drag_at: 'Dragged',
focus: 'Focused element',
fill: 'Filled field',
clear: 'Cleared field',
check: 'Checked box',
uncheck: 'Unchecked box',
press_key: 'Pressed key',
upload_file: 'Uploaded file',
// Console / scripts
evaluate_script: 'Ran script',
get_console_logs: 'Read console logs',
// History / bookmarks
search_history: 'Searched history',
get_recent_history: 'Read recent history',
delete_history_url: 'Deleted history entry',
delete_history_range: 'Deleted history range',
get_bookmarks: 'Listed bookmarks',
create_bookmark: 'Created bookmark',
remove_bookmark: 'Removed bookmark',
update_bookmark: 'Updated bookmark',
move_bookmark: 'Moved bookmark',
search_bookmarks: 'Searched bookmarks',
// Filesystem (sandboxed)
read_file: 'Read file',
write_file: 'Wrote file',
find_files: 'Searched files',
// Memory
read_soul: 'Read soul memory',
read_core: 'Read core memory',
write_memory: 'Wrote memory',
search_memory: 'Searched memory',
update_soul: 'Updated soul memory',
update_core: 'Updated core memory',
// Web
web_search: 'Searched the web',
web_fetch: 'Fetched URL',
// Klavis / external apps (Strata)
connector_mcp_servers: 'Listed connected apps',
discover_server_categories_or_actions: 'Browsed available actions',
get_category_actions: 'Listed actions',
get_action_details: 'Looked up action',
execute_action: 'Ran external action',
search_documentation: 'Searched docs',
handle_auth_failure: 'Handled auth issue',
// Suggestions
suggest_schedule: 'Suggested schedule',
suggest_app_connection: 'Suggested app connect',
// BrowserOS info
browseros_info: 'Read BrowserOS info',
// Windows
list_windows: 'Listed windows',
focus_window: 'Focused window',
close_window: 'Closed window',
create_window: 'Created window',
}
// ──────────────────────────────────────────────────────────────────────
// Helpers
// ──────────────────────────────────────────────────────────────────────
function asString(value: unknown): string | undefined {
return typeof value === 'string' && value.length > 0 ? value : undefined
}
function stringField(
input: Record<string, unknown>,
...keys: string[]
): string | undefined {
for (const k of keys) {
const v = asString(input[k])
if (v) return v
}
return undefined
}
function truncate(text: string | undefined, max: number): string | undefined {
if (!text) return undefined
return text.length > max ? `${text.slice(0, max - 1)}` : text
}
function quote(value: string | undefined): string | undefined {
if (!value) return undefined
return `"${truncate(value, 60)}"`
}
function basename(path: string | undefined): string | undefined {
if (!path) return undefined
const parts = path.split(/[/\\]/).filter(Boolean)
return parts[parts.length - 1] ?? path
}
function formatUrl(value: unknown): string | undefined {
const url = asString(value)
if (!url) return undefined
try {
const parsed = new URL(url)
const host = parsed.host
const path = parsed.pathname === '/' ? '' : parsed.pathname
const display = path && path.length > 0 ? `${host}${path}` : host
return truncate(display, 60)
} catch {
return truncate(url, 60)
}
}
function coords(x: unknown, y: unknown): string | undefined {
if (typeof x === 'number' && typeof y === 'number') {
return `${Math.round(x)}, ${Math.round(y)}`
}
return undefined
}
// ──────────────────────────────────────────────────────────────────────
// Subject extractors
// ──────────────────────────────────────────────────────────────────────
type SubjectExtractor = (input: Record<string, unknown>) => string | undefined
const SUBJECT_EXTRACTORS: Record<string, SubjectExtractor> = {
// URL-bearing tools
new_page: (i) => formatUrl(i.url),
new_hidden_page: (i) => formatUrl(i.url),
navigate_page: (i) => {
const action = asString(i.action)
if (action === 'back') return 'back'
if (action === 'forward') return 'forward'
if (action === 'reload') return 'reload'
return formatUrl(i.url)
},
web_fetch: (i) => formatUrl(i.url),
// Search queries
web_search: (i) => quote(stringField(i, 'query', 'q')),
search_history: (i) => quote(stringField(i, 'query', 'text')),
search_bookmarks: (i) => quote(stringField(i, 'query', 'text')),
search_memory: (i) => quote(stringField(i, 'query', 'q')),
search_dom: (i) => quote(stringField(i, 'query', 'selector')),
search_documentation: (i) => quote(stringField(i, 'query', 'q')),
find_files: (i) => quote(stringField(i, 'pattern', 'query')),
// Element interactions
click: (i) => stringField(i, 'element'),
hover: (i) => stringField(i, 'element'),
focus: (i) => stringField(i, 'element'),
clear: (i) => stringField(i, 'element'),
check: (i) => stringField(i, 'element'),
uncheck: (i) => stringField(i, 'element'),
fill: (i) => {
const target = stringField(i, 'element')
const text = stringField(i, 'text')
if (target && text) return `${target}: ${truncate(text, 40)}`
return target ?? truncate(text, 40)
},
press_key: (i) => stringField(i, 'key'),
// Coordinate-based input
click_at: (i) => coords(i.x, i.y),
hover_at: (i) => coords(i.x, i.y),
type_at: (i) => {
const at = coords(i.x, i.y)
const text = stringField(i, 'text')
if (at && text) return `${at}: ${truncate(text, 40)}`
return at ?? truncate(text, 40)
},
drag_at: (i) => {
const from = coords(i.fromX, i.fromY)
const to = coords(i.toX, i.toY)
if (from && to) return `${from}${to}`
return from ?? to
},
// Tab management
show_page: (i) => {
const page = i.page
return typeof page === 'number' ? `tab ${page}` : asString(page)
},
close_page: (i) => {
const page = i.page
return typeof page === 'number' ? `tab ${page}` : asString(page)
},
move_page: (i) => {
const page = i.page
return typeof page === 'number' ? `tab ${page}` : asString(page)
},
// Page reads (take_snapshot, take_enhanced_snapshot, get_page_content,
// get_page_links, get_dom, take_screenshot) intentionally omit a
// subject — the only argument is a numeric page ID that's internal
// to the agent and meaningless to the user ("tab 4" tells them nothing).
// The verb alone communicates what happened.
// External actions via Strata
execute_action: (i) => {
const server = stringField(i, 'server_name')
const action = stringField(i, 'action_name')
if (server && action) return `${server} · ${action}`
return action ?? server
},
get_category_actions: (i) => stringField(i, 'category_name', 'server_name'),
get_action_details: (i) => stringField(i, 'action_name'),
discover_server_categories_or_actions: (i) =>
stringField(i, 'server_name', 'category_name'),
connector_mcp_servers: (i) => stringField(i, 'server_name'),
// Filesystem
read_file: (i) => basename(stringField(i, 'path')),
write_file: (i) => basename(stringField(i, 'path')),
// Memory writes — show first chars of content
write_memory: (i) => truncate(stringField(i, 'content', 'text'), 40),
update_soul: (i) => truncate(stringField(i, 'content'), 40),
update_core: (i) => truncate(stringField(i, 'content'), 40),
// Bookmarks
create_bookmark: (i) => stringField(i, 'title') ?? formatUrl(i.url),
remove_bookmark: (i) => stringField(i, 'id', 'title'),
update_bookmark: (i) => stringField(i, 'id', 'title'),
move_bookmark: (i) => stringField(i, 'id', 'title'),
// History
delete_history_url: (i) => formatUrl(i.url),
}
// ──────────────────────────────────────────────────────────────────────
// Public API
// ──────────────────────────────────────────────────────────────────────
export interface ToolLabelResult {
label: string
subject?: string
}
/**
* Strip MCP namespace prefixes (e.g. "browseros__", "mcp_") to find the
* canonical tool name used in the override maps.
*/
function canonicalName(rawName: string): string {
return rawName.replace(/^browseros__/, '').replace(/^mcp_/, '')
}
/**
* Convert a snake_case tool name into Sentence-case English as a fallback
* when no curated override exists.
*/
function humanizeToolName(rawName: string): string {
const stripped = canonicalName(rawName)
const words = stripped.split(/[_-]/).filter((w) => w.length > 0)
if (words.length === 0) return rawName
const first = words[0]!
return [
first.charAt(0).toUpperCase() + first.slice(1),
...words.slice(1),
].join(' ')
}
/**
* Build a human-readable label and subject string for a tool call,
* suitable for rendering in the chat activity view.
*/
export function buildToolLabel(
rawName: string,
input?: Record<string, unknown>,
): ToolLabelResult {
const canonical = canonicalName(rawName)
const label =
VERB_OVERRIDES[canonical] ??
VERB_OVERRIDES[rawName] ??
humanizeToolName(rawName)
const extractor = Object.hasOwn(SUBJECT_EXTRACTORS, canonical)
? SUBJECT_EXTRACTORS[canonical]
: Object.hasOwn(SUBJECT_EXTRACTORS, rawName)
? SUBJECT_EXTRACTORS[rawName]
: undefined
const subject = extractor && input ? extractor(input) : undefined
return { label, subject }
}

View File

@@ -7,8 +7,9 @@ import { PRODUCT_WEB_HOST } from './lib/constants/productWebHost'
// biome-ignore lint/style/noProcessEnv: build config file needs env access
const env = process.env
// biome-ignore lint/style/noNonNullAssertion: required env var
const apiUrl = new URL(env.VITE_PUBLIC_BROWSEROS_API!)
const apiUrl = new URL(
env.VITE_PUBLIC_BROWSEROS_API?.trim() || 'https://api.browseros.com',
)
const apiPattern = apiUrl.port
? `${apiUrl.hostname}:${apiUrl.port}`
: apiUrl.hostname

View File

@@ -0,0 +1,26 @@
{
"agent": {
"type": "single",
"provider": "openai-compatible",
"model": "moonshotai/kimi-k2.5",
"apiKey": "OPENROUTER_API_KEY",
"baseUrl": "https://openrouter.ai/api/v1",
"supportsImages": true
},
"dataset": "../data/agisdk-real.jsonl",
"num_workers": 10,
"restart_server_per_task": true,
"browseros": {
"server_url": "http://127.0.0.1:9110",
"base_cdp_port": 9010,
"base_server_port": 9110,
"base_extension_port": 9310,
"load_extensions": false,
"headless": false
},
"captcha": {
"api_key_env": "NOPECHA_API_KEY"
},
"graders": ["agisdk_state_diff"],
"timeout_ms": 1800000
}

View File

@@ -2,9 +2,9 @@
"agent": {
"type": "single",
"provider": "openai-compatible",
"model": "accounts/fireworks/models/kimi-k2p5",
"apiKey": "FIREWORKS_API_KEY",
"baseUrl": "https://api.fireworks.ai/inference/v1",
"model": "moonshotai/kimi-k2.5",
"apiKey": "OPENROUTER_API_KEY",
"baseUrl": "https://openrouter.ai/api/v1",
"supportsImages": true
},
"dataset": "../data/webbench-2of4-50.jsonl",

View File

@@ -0,0 +1,26 @@
{
"agent": {
"type": "single",
"provider": "openai-compatible",
"model": "moonshotai/kimi-k2.5",
"apiKey": "OPENROUTER_API_KEY",
"baseUrl": "https://openrouter.ai/api/v1",
"supportsImages": true
},
"dataset": "../data/webarena-infinity-hard-50.jsonl",
"num_workers": 10,
"restart_server_per_task": true,
"browseros": {
"server_url": "http://127.0.0.1:9110",
"base_cdp_port": 9010,
"base_server_port": 9110,
"base_extension_port": 9310,
"load_extensions": false,
"headless": false
},
"captcha": {
"api_key_env": "NOPECHA_API_KEY"
},
"graders": ["infinity_state"],
"timeout_ms": 1800000
}

View File

@@ -0,0 +1,47 @@
{"query_id": "agisdk-dashdish-10", "dataset": "agisdk-real", "query": "Place an order from \"Souvla\" for a \"Medium Classic Cheeseburger\" and a \"Small Bacon Double Cheeseburger\" with \"Standard Delivery\" as the method with the default charged options.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-dashdish.vercel.app", "metadata": {"original_task_id": "dashdish-10", "website": "DashDish", "category": "agisdk-real", "additional": {"agisdk_task_id": "dashdish-10", "challenge_type": "action", "difficulty": "hard", "similar_to": "Doordash"}}}
{"query_id": "agisdk-fly-unified-5", "dataset": "agisdk-real", "query": "Find me the cheapest fare for a flight from Orlando to Milwaukee on December 5th, 2024 and book it.\nPassenger: John Doe\nDate of Birth: 01/01/1990\nSex: Male\nSeat Selection: No\nPayment: Credit Card (378342143523967), Exp: 12/25, Security Code: 420 Address: 123 Main St, San Francisco, CA, 94105, USA, Phone: 555-123-4567, Email: johndoe@example.com.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-fly-unified.vercel.app", "metadata": {"original_task_id": "fly-unified-5", "website": "Fly Unified", "category": "agisdk-real", "additional": {"agisdk_task_id": "fly-unified-5", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "United Airlines"}}}
{"query_id": "agisdk-udriver-10", "dataset": "agisdk-real", "query": "Order me a ride for 4pm, I'll be at the de Young muesum headed to the Waterbar, fanciest option possible please.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-udriver.vercel.app", "metadata": {"original_task_id": "udriver-10", "website": "UDriver", "category": "agisdk-real", "additional": {"agisdk_task_id": "udriver-10", "challenge_type": "action", "difficulty": "hard", "similar_to": "Uber"}}}
{"query_id": "agisdk-udriver-9", "dataset": "agisdk-real", "query": "Book me a ride from the thai restaurant I last took a ride to for later today at 2pm, I'll be at 333 Apartments on Fremont", "graders": ["agisdk_state_diff"], "start_url": "https://evals-udriver.vercel.app", "metadata": {"original_task_id": "udriver-9", "website": "UDriver", "category": "agisdk-real", "additional": {"agisdk_task_id": "udriver-9", "challenge_type": "retrieval-action", "difficulty": "hard", "similar_to": "Uber"}}}
{"query_id": "agisdk-topwork-4", "dataset": "agisdk-real", "query": "Create a job post for a UI/UX Designer with expertise in Figma, Sketch, and Adobe Creative Suite, including project details, timeline, and required skills (Wireframing, Prototyping, Responsive Design).", "graders": ["agisdk_state_diff"], "start_url": "https://evals-topwork.vercel.app", "metadata": {"original_task_id": "topwork-4", "website": "TopWork", "category": "agisdk-real", "additional": {"agisdk_task_id": "topwork-4", "challenge_type": "action", "difficulty": "medium", "similar_to": "Upwork"}}}
{"query_id": "agisdk-gocalendar-4", "dataset": "agisdk-real", "query": "Change the \"Team Check-In\" event on July 18, 2024, name to \"Project Kickoff\" and update the location to \"Zoom\"", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gocalendar.vercel.app", "metadata": {"original_task_id": "gocalendar-4", "website": "GoCalendar", "category": "agisdk-real", "additional": {"agisdk_task_id": "gocalendar-4", "challenge_type": "action", "difficulty": "medium", "similar_to": "Google Calendar"}}}
{"query_id": "agisdk-staynb-6", "dataset": "agisdk-real", "query": "Find and book the stay with the best value for money (cheapest stay with the best reviews) for 1 day. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-staynb.vercel.app", "metadata": {"original_task_id": "staynb-6", "website": "StayNB", "category": "agisdk-real", "additional": {"agisdk_task_id": "staynb-6", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "Airbnb"}}}
{"query_id": "agisdk-fly-unified-9", "dataset": "agisdk-real", "query": "Book me a flight from San Francisco to Chicago in Basic Economy on December 18th at 10:00. Ensure no seat selection is made.\nPassenger: David Lee\nDate of Birth: 07/22/1985\nSex: Male\nSeat Selection: No\nPayment: Credit Card (9999 8888 7777), Exp: 03/30, Address: 987 Cedar St, Chicago, IL, 60601, USA, Phone: 555-987-1234, Email: davidlee@example.com.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-fly-unified.vercel.app", "metadata": {"original_task_id": "fly-unified-9", "website": "Fly Unified", "category": "agisdk-real", "additional": {"agisdk_task_id": "fly-unified-9", "challenge_type": "action", "difficulty": "hard", "similar_to": "United Airlines"}}}
{"query_id": "agisdk-networkin-9", "dataset": "agisdk-real", "query": "Find a professional who attended Stanford and send them a connection request and a message.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-networkin.vercel.app", "metadata": {"original_task_id": "networkin-9", "website": "Networkin", "category": "agisdk-real", "additional": {"agisdk_task_id": "networkin-9", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "LinkedIn"}}}
{"query_id": "agisdk-udriver-11", "dataset": "agisdk-real", "query": "I need to go from Pacific Catch on Chestnut back home to 333 Fremont now. If the fancy version is within ten dollars of the regular one, book that.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-udriver.vercel.app", "metadata": {"original_task_id": "udriver-11", "website": "UDriver", "category": "agisdk-real", "additional": {"agisdk_task_id": "udriver-11", "challenge_type": "action", "difficulty": "hard", "similar_to": "Uber"}}}
{"query_id": "agisdk-fly-unified-4", "dataset": "agisdk-real", "query": "Book me a round-trip flight from Providence (Rhode Island) to Indianapolis, departing on December 5th, 2024 at 08:00 and returning on December 9th at 14:00.\nPassenger: Jane Smith\nDate of Birth: 02/14/1995\nSex: Female\nSeat Selection: Yes (Window seat)\nPayment: Credit Card (378342143523967), Exp: 06/26, security code: 345 Address: 456 Elm St, Miami, FL, 33101, USA, Phone: 555-987-6543, Email: janesmith@example.com.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-fly-unified.vercel.app", "metadata": {"original_task_id": "fly-unified-4", "website": "Fly Unified", "category": "agisdk-real", "additional": {"agisdk_task_id": "fly-unified-4", "challenge_type": "action", "difficulty": "medium", "similar_to": "United Airlines"}}}
{"query_id": "agisdk-networkin-5", "dataset": "agisdk-real", "query": "Send a connection request to John Smith.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-networkin.vercel.app", "metadata": {"original_task_id": "networkin-5", "website": "Networkin", "category": "agisdk-real", "additional": {"agisdk_task_id": "networkin-5", "challenge_type": "action", "difficulty": "easy", "similar_to": "LinkedIn"}}}
{"query_id": "agisdk-zilloft-6", "dataset": "agisdk-real", "query": "Select a property listed in San Francisco as \"Condos\" within a price range under $300,000 and request a tour for tomorrow at 4:00 PM. Use these contact details: Name: Sarah Brown, Email: sarahbrown@example.com, Phone: 555-987-6543.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-zilloft.vercel.app", "metadata": {"original_task_id": "zilloft-6", "website": "Zilloft", "category": "agisdk-real", "additional": {"agisdk_task_id": "zilloft-6", "challenge_type": "action", "difficulty": "medium", "similar_to": "Zillow"}}}
{"query_id": "agisdk-topwork-2", "dataset": "agisdk-real", "query": "Create a job posting for a Backend Developer specializing in Python, Django, and Flask to develop a high-performance web application. Include project details such as required skills (PostgreSQL, Docker, AWS, CI/CD), estimated project timeline, and budget.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-topwork.vercel.app", "metadata": {"original_task_id": "topwork-2", "website": "TopWork", "category": "agisdk-real", "additional": {"agisdk_task_id": "topwork-2", "challenge_type": "action", "difficulty": "medium", "similar_to": "Upwork"}}}
{"query_id": "agisdk-gocalendar-3", "dataset": "agisdk-real", "query": "Delete the event titled \"Breakfast Meeting with Client\" scheduled for July 19, 2024", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gocalendar.vercel.app", "metadata": {"original_task_id": "gocalendar-3", "website": "GoCalendar", "category": "agisdk-real", "additional": {"agisdk_task_id": "gocalendar-3", "challenge_type": "action", "difficulty": "easy", "similar_to": "Google Calendar"}}}
{"query_id": "agisdk-topwork-3", "dataset": "agisdk-real", "query": "Create a job listing for a Full-Stack Developer with expertise in Java, Spring Boot, and Angular, outlining the project scope, estimated duration, and required skills (MySQL, Docker, Kubernetes, and Jenkins). The ideal candidate should have experience in enterprise-level applications and building scalable microservices. After creating the job post, please describe what you included in the job listing.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-topwork.vercel.app", "metadata": {"original_task_id": "topwork-3", "website": "TopWork", "category": "agisdk-real", "additional": {"agisdk_task_id": "topwork-3", "challenge_type": "retrieval", "difficulty": "medium", "similar_to": "Upwork"}}}
{"query_id": "agisdk-fly-unified-2", "dataset": "agisdk-real", "query": "Book me a one-way flight from Indiana to New York on December 2nd 2024 at 12:00.\nPassenger: John Doe\nDate of Birth: 01/01/1990\nSex: Male\nSeat Selection: No\nPayment: Credit Card (378342143523967), Exp: 12/25, Security Code: 245, Address: 123 Main St, San Francisco, CA, 94105, USA, Phone: 555-123-4567, Email: johndoe@example.com.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-fly-unified.vercel.app", "metadata": {"original_task_id": "fly-unified-2", "website": "Fly Unified", "category": "agisdk-real", "additional": {"agisdk_task_id": "fly-unified-2", "challenge_type": "action", "difficulty": "easy", "similar_to": "United Airlines"}}}
{"query_id": "agisdk-dashdish-7", "dataset": "agisdk-real", "query": "Select \"Express Delivery\" for an order from \"DragonEats\" of \"Mushroom Swiss Burger\" and complete the checkout with the pre-loaded Visa card.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-dashdish.vercel.app", "metadata": {"original_task_id": "dashdish-7", "website": "DashDish", "category": "agisdk-real", "additional": {"agisdk_task_id": "dashdish-7", "challenge_type": "action", "difficulty": "hard", "similar_to": "Doordash"}}}
{"query_id": "agisdk-networkin-3", "dataset": "agisdk-real", "query": "Write a post inviting users to a networking event, including details about the event's purpose, date, and target audience.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-networkin.vercel.app", "metadata": {"original_task_id": "networkin-3", "website": "Networkin", "category": "agisdk-real", "additional": {"agisdk_task_id": "networkin-3", "challenge_type": "action", "difficulty": "medium", "similar_to": "LinkedIn"}}}
{"query_id": "agisdk-gomail-7", "dataset": "agisdk-real", "query": "Delete the email with the subject \"New Leadership Articles You Can't Miss\" from the Inbox.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gomail.vercel.app", "metadata": {"original_task_id": "gomail-7", "website": "GoMail", "category": "agisdk-real", "additional": {"agisdk_task_id": "gomail-7", "challenge_type": "retrieval-action", "difficulty": "hard", "similar_to": "Gmail"}}}
{"query_id": "agisdk-opendining-8", "dataset": "agisdk-real", "query": "Identify and book the restaurant with the lowest rating. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-opendining.vercel.app", "metadata": {"original_task_id": "opendining-8", "website": "OpenDining", "category": "agisdk-real", "additional": {"agisdk_task_id": "opendining-8", "challenge_type": "retrieval-action", "difficulty": "easy", "similar_to": "OpenTable"}}}
{"query_id": "agisdk-udriver-1", "dataset": "agisdk-real", "query": "Book a ride from Fitness Urbano to Pacific Cafe", "graders": ["agisdk_state_diff"], "start_url": "https://evals-udriver.vercel.app", "metadata": {"original_task_id": "udriver-1", "website": "UDriver", "category": "agisdk-real", "additional": {"agisdk_task_id": "udriver-1", "challenge_type": "action", "difficulty": "easy", "similar_to": "Uber"}}}
{"query_id": "agisdk-staynb-2", "dataset": "agisdk-real", "query": "Click on one of the stays displayed on the homepage and book it for a family of 4 (2 adults and 2 children). For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-staynb.vercel.app", "metadata": {"original_task_id": "staynb-2", "website": "StayNB", "category": "agisdk-real", "additional": {"agisdk_task_id": "staynb-2", "challenge_type": "action", "difficulty": "easy", "similar_to": "Airbnb"}}}
{"query_id": "agisdk-opendining-10", "dataset": "agisdk-real", "query": "Check the menus of all restaurants for vegetarian options and make a reservation at the one with the most vegetarian choices. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-opendining.vercel.app", "metadata": {"original_task_id": "opendining-10", "website": "OpenDining", "category": "agisdk-real", "additional": {"agisdk_task_id": "opendining-10", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "OpenTable"}}}
{"query_id": "agisdk-opendining-4", "dataset": "agisdk-real", "query": "Use the search bar to search for a restaurant on September 2nd at 4:30 PM for 7 people, using \"Japanese\" as the search term, and book the first result. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-opendining.vercel.app", "metadata": {"original_task_id": "opendining-4", "website": "OpenDining", "category": "agisdk-real", "additional": {"agisdk_task_id": "opendining-4", "challenge_type": "action", "difficulty": "hard", "similar_to": "OpenTable"}}}
{"query_id": "agisdk-gomail-8", "dataset": "agisdk-real", "query": "Clear all emails from \"GitHub\" in the inbox to trash.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gomail.vercel.app", "metadata": {"original_task_id": "gomail-8", "website": "GoMail", "category": "agisdk-real", "additional": {"agisdk_task_id": "gomail-8", "challenge_type": "action", "difficulty": "medium", "similar_to": "Gmail"}}}
{"query_id": "agisdk-dashdish-4", "dataset": "agisdk-real", "query": "Schedule a delivery order from \"Taco Bell\" adding a \"Classic Cheeseburger\" large size for later and add the note \"Leave at the front door\".", "graders": ["agisdk_state_diff"], "start_url": "https://evals-dashdish.vercel.app", "metadata": {"original_task_id": "dashdish-4", "website": "DashDish", "category": "agisdk-real", "additional": {"agisdk_task_id": "dashdish-4", "challenge_type": "action", "difficulty": "medium", "similar_to": "Doordash"}}}
{"query_id": "agisdk-networkin-1", "dataset": "agisdk-real", "query": "Create a new text post for the feed with a professional update about AI trends in 2025, mentioning three key advancements and their impact on the job market.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-networkin.vercel.app", "metadata": {"original_task_id": "networkin-1", "website": "Networkin", "category": "agisdk-real", "additional": {"agisdk_task_id": "networkin-1", "challenge_type": "action", "difficulty": "medium", "similar_to": "LinkedIn"}}}
{"query_id": "agisdk-dashdish-5", "dataset": "agisdk-real", "query": "Add three \"Loaded Bacon Cheese Fries\" to the shopping cart from \"Man vs. Fries\". Proceed to checkout and select \"Pickup\" as the delivery method.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-dashdish.vercel.app", "metadata": {"original_task_id": "dashdish-5", "website": "DashDish", "category": "agisdk-real", "additional": {"agisdk_task_id": "dashdish-5", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "Doordash"}}}
{"query_id": "agisdk-opendining-5", "dataset": "agisdk-real", "query": "Scroll through the homepage carousel until \"Ocean Breeze\" is visible, select the second available time slot, and complete the reservation. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-opendining.vercel.app", "metadata": {"original_task_id": "opendining-5", "website": "OpenDining", "category": "agisdk-real", "additional": {"agisdk_task_id": "opendining-5", "challenge_type": "action", "difficulty": "medium", "similar_to": "OpenTable"}}}
{"query_id": "agisdk-topwork-1", "dataset": "agisdk-real", "query": "Create a new job post for a Frontend Developer with expertise in React and TypeScript, specifying project details such as estimated duration, required skills, and budget.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-topwork.vercel.app", "metadata": {"original_task_id": "topwork-1", "website": "TopWork", "category": "agisdk-real", "additional": {"agisdk_task_id": "topwork-1", "challenge_type": "action", "difficulty": "medium", "similar_to": "Upwork"}}}
{"query_id": "agisdk-gocalendar-1", "dataset": "agisdk-real", "query": "Create a new event titled \"Team Meeting\" on July 19, 2024, from 2 PM to 2:30 PM, and include \"Conference Room A\" as the location", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gocalendar.vercel.app", "metadata": {"original_task_id": "gocalendar-1", "website": "GoCalendar", "category": "agisdk-real", "additional": {"agisdk_task_id": "gocalendar-1", "challenge_type": "action", "difficulty": "medium", "similar_to": "Google Calendar"}}}
{"query_id": "agisdk-gomail-5", "dataset": "agisdk-real", "query": "Schedule an email to jane.doe@example.com with the subject \"Weekly Update\" to be sent next Monday at 9:00 AM.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gomail.vercel.app", "metadata": {"original_task_id": "gomail-5", "website": "GoMail", "category": "agisdk-real", "additional": {"agisdk_task_id": "gomail-5", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "Gmail"}}}
{"query_id": "agisdk-staynb-4", "dataset": "agisdk-real", "query": "Book a stay for 2 children with 1 adult. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-staynb.vercel.app", "metadata": {"original_task_id": "staynb-4", "website": "StayNB", "category": "agisdk-real", "additional": {"agisdk_task_id": "staynb-4", "challenge_type": "action", "difficulty": "medium", "similar_to": "Airbnb"}}}
{"query_id": "agisdk-networkin-6", "dataset": "agisdk-real", "query": "Choose a random person who you haven't connected with, connect with them, and send them a message saying, 'howdy, partner'.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-networkin.vercel.app", "metadata": {"original_task_id": "networkin-6", "website": "Networkin", "category": "agisdk-real", "additional": {"agisdk_task_id": "networkin-6", "challenge_type": "action", "difficulty": "medium", "similar_to": "LinkedIn"}}}
{"query_id": "agisdk-dashdish-2", "dataset": "agisdk-real", "query": "Add a \"Medium Pepperoni Pizza\" from the restaurant \"Papa Johns Pizza\" to the shopping cart and purchase it.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-dashdish.vercel.app", "metadata": {"original_task_id": "dashdish-2", "website": "DashDish", "category": "agisdk-real", "additional": {"agisdk_task_id": "dashdish-2", "challenge_type": "action", "difficulty": "easy", "similar_to": "Doordash"}}}
{"query_id": "agisdk-staynb-8", "dataset": "agisdk-real", "query": "Scroll through the homepage and book the last stay located in Paris.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-staynb.vercel.app", "metadata": {"original_task_id": "staynb-8", "website": "StayNB", "category": "agisdk-real", "additional": {"agisdk_task_id": "staynb-8", "challenge_type": "retrieval-action", "difficulty": "medium", "similar_to": "Airbnb"}}}
{"query_id": "agisdk-gomail-2", "dataset": "agisdk-real", "query": "Mark the first email in the Inbox as \"read\".", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gomail.vercel.app", "metadata": {"original_task_id": "gomail-2", "website": "GoMail", "category": "agisdk-real", "additional": {"agisdk_task_id": "gomail-2", "challenge_type": "action", "difficulty": "easy", "similar_to": "Gmail"}}}
{"query_id": "agisdk-networkin-10", "dataset": "agisdk-real", "query": "Generate a polite follow-up message for a previous unanswered chat, starting with \"Following up on\".", "graders": ["agisdk_state_diff"], "start_url": "https://evals-networkin.vercel.app", "metadata": {"original_task_id": "networkin-10", "website": "Networkin", "category": "agisdk-real", "additional": {"agisdk_task_id": "networkin-10", "challenge_type": "action", "difficulty": "medium", "similar_to": "LinkedIn"}}}
{"query_id": "agisdk-gomail-3", "dataset": "agisdk-real", "query": "Compose a new email to jonathan.smith@example.com with the subject \"Meeting Notes\" and body \"Please find the meeting notes attached.\"", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gomail.vercel.app", "metadata": {"original_task_id": "gomail-3", "website": "GoMail", "category": "agisdk-real", "additional": {"agisdk_task_id": "gomail-3", "challenge_type": "action", "difficulty": "easy", "similar_to": "Gmail"}}}
{"query_id": "agisdk-udriver-6", "dataset": "agisdk-real", "query": "Me and 4 friends need a ride from the Palace Hotel to dinner at Osha Thai leaving now", "graders": ["agisdk_state_diff"], "start_url": "https://evals-udriver.vercel.app", "metadata": {"original_task_id": "udriver-6", "website": "UDriver", "category": "agisdk-real", "additional": {"agisdk_task_id": "udriver-6", "challenge_type": "action", "difficulty": "hard", "similar_to": "Uber"}}}
{"query_id": "agisdk-staynb-9", "dataset": "agisdk-real", "query": "Book a stay with the maximum number of guests supported. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-staynb.vercel.app", "metadata": {"original_task_id": "staynb-9", "website": "StayNB", "category": "agisdk-real", "additional": {"agisdk_task_id": "staynb-9", "challenge_type": "action", "difficulty": "hard", "similar_to": "Airbnb"}}}
{"query_id": "agisdk-zilloft-3", "dataset": "agisdk-real", "query": "Find a home in San Diego priced under $150,000 with at least 2 bedrooms and request a tour. Use these details: Contact Name: John Doe, Email: johndoe@example.com, Phone: 555-123-4567, Tour Time: 2:00 PM, Tour Date: First available.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-zilloft.vercel.app", "metadata": {"original_task_id": "zilloft-3", "website": "Zilloft", "category": "agisdk-real", "additional": {"agisdk_task_id": "zilloft-3", "challenge_type": "retrieval-action", "difficulty": "easy", "similar_to": "Zillow"}}}
{"query_id": "agisdk-fly-unified-6", "dataset": "agisdk-real", "query": "Reserve me a seat for the flight from Austin to Pittsburgh departing on December 11th, 2024 at 8:00 in Basic Economy.\nPassenger: Alice Brown\nDate of Birth: 05/20/1992\nSex: Female\nSeat Selection: Yes (Aisle seat)\nPayment: Credit Card (378342143523967), Exp: 09/27, security code: 332 Address: 789 Pine St, Los Angeles, CA, 90012, USA, Phone: 555-456-7890, Email: alicebrown@example.com.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-fly-unified.vercel.app", "metadata": {"original_task_id": "fly-unified-6", "website": "Fly Unified", "category": "agisdk-real", "additional": {"agisdk_task_id": "fly-unified-6", "challenge_type": "action", "difficulty": "medium", "similar_to": "United Airlines"}}}
{"query_id": "agisdk-opendining-3", "dataset": "agisdk-real", "query": "Book a table at \"The Royal Dine\" for a party of 4 on July 20, 2024, at 7 PM. For fields you don't know the answer for, just fill them in with anything of your choice.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-opendining.vercel.app", "metadata": {"original_task_id": "opendining-3", "website": "OpenDining", "category": "agisdk-real", "additional": {"agisdk_task_id": "opendining-3", "challenge_type": "action", "difficulty": "easy", "similar_to": "OpenTable"}}}
{"query_id": "agisdk-gocalendar-7", "dataset": "agisdk-real", "query": "Reschedule the \"Morning Coffee with sister\" event from July 18, 2024, at 9 AM to July 19, 2024, at 10AM using drag-and-drop functionality", "graders": ["agisdk_state_diff"], "start_url": "https://evals-gocalendar.vercel.app", "metadata": {"original_task_id": "gocalendar-7", "website": "GoCalendar", "category": "agisdk-real", "additional": {"agisdk_task_id": "gocalendar-7", "challenge_type": "action", "difficulty": "medium", "similar_to": "Google Calendar"}}}
{"query_id": "agisdk-staynb-5", "dataset": "agisdk-real", "query": "Use the search bar to look for a stay. For the \"Where\" section, use the \"Search by region\" popover and select \"Europe\". Set the check-in date to October 13th and the check-out date to October 23rd. For the \"Who\" section, select 1 infant, 2 children, and 2 adults. Press the search button, select the first stay, and book it.", "graders": ["agisdk_state_diff"], "start_url": "https://evals-staynb.vercel.app", "metadata": {"original_task_id": "staynb-5", "website": "StayNB", "category": "agisdk-real", "additional": {"agisdk_task_id": "staynb-5", "challenge_type": "action", "difficulty": "medium", "similar_to": "Airbnb"}}}

View File

@@ -0,0 +1,50 @@
{"query_id": "infinity-elation-prescriptions-task_h69", "dataset": "webarena-infinity", "query": "Approve all pending refill requests except for any medication that is involved in a major drug-drug interaction with another of the patient's active medications. Deny those with the reason 'Drug interaction \u2014 needs provider review before renewal'.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h69", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h69.py", "app_base_port": 8020}}}
{"query_id": "infinity-elation-clinical-records-task_h52", "dataset": "webarena-infinity", "query": "Add the document tag 'Provider-Reviewed' to every visit note template that was created by the current logged-in provider. Do not modify templates created by other providers.", "graders": ["infinity_state"], "start_url": "http://localhost:8000", "metadata": {"original_task_id": "elation-clinical-records-task_h52", "website": "elation-clinical-records", "category": "webarena-infinity", "additional": {"app_name": "elation-clinical-records", "difficulty": "hard", "verifier_path": "real-tasks/task_h52.py", "app_base_port": 8000}}}
{"query_id": "infinity-gmail-accounts-and-contacts-task_h44", "dataset": "webarena-infinity", "query": "Your sister's husband is one of your contacts. Find him, star his entry, and add the Friends label.", "graders": ["infinity_state"], "start_url": "http://localhost:8070", "metadata": {"original_task_id": "gmail-accounts-and-contacts-task_h44", "website": "gmail-accounts-and-contacts", "category": "webarena-infinity", "additional": {"app_name": "gmail-accounts-and-contacts", "difficulty": "hard", "verifier_path": "real-tasks/task_h44.py", "app_base_port": 8070}}}
{"query_id": "infinity-gmail-task_h2", "dataset": "webarena-infinity", "query": "Update the Datadog alerts filter to also archive matching emails and forward them to priya.sharma@cloudnine.dev instead of nate.patel@devops.tools.", "graders": ["infinity_state"], "start_url": "http://localhost:8060", "metadata": {"original_task_id": "gmail-task_h2", "website": "gmail", "category": "webarena-infinity", "additional": {"app_name": "gmail", "difficulty": "hard", "verifier_path": "real-tasks/task_h2.py", "app_base_port": 8060}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h58", "dataset": "webarena-infinity", "query": "The Performance Initiative epic has two child epics. For the child epic with more open issues, set the weight of every issue in it to 13. For the other child epic, close all its open issues.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h58", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h58.py", "app_base_port": 8050}}}
{"query_id": "infinity-figma-slides-task_h46", "dataset": "webarena-infinity", "query": "There are two slides with tables in the deck. Lock the table that compares competitors, and change the font size to 16 on the table that tracks quarterly feature adoption.", "graders": ["infinity_state"], "start_url": "http://localhost:8030", "metadata": {"original_task_id": "figma-slides-task_h46", "website": "figma-slides", "category": "webarena-infinity", "additional": {"app_name": "figma-slides", "difficulty": "hard", "verifier_path": "real-tasks/task_h46.py", "app_base_port": 8030}}}
{"query_id": "infinity-elation-prescriptions-task_h50", "dataset": "webarena-infinity", "query": "Deny the pending refill for the patient's cholesterol medication because his lipid panel is overdue. Then deny the Lisinopril refill as well \u2014 he needs a follow-up blood pressure check first.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h50", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h50.py", "app_base_port": 8020}}}
{"query_id": "infinity-elation-prescriptions-task_h19", "dataset": "webarena-infinity", "query": "Discontinue the Omeprazole and prescribe Famotidine 20mg tablet twice daily as a replacement for GERD \u2014 qty 60, 3 refills, send to CVS #4521.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h19", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h19.py", "app_base_port": 8020}}}
{"query_id": "infinity-paypal-my-wallet-task_h25", "dataset": "webarena-infinity", "query": "Convert all of my Australian dollars to euros.", "graders": ["infinity_state"], "start_url": "http://localhost:8100", "metadata": {"original_task_id": "paypal-my-wallet-task_h25", "website": "paypal-my-wallet", "category": "webarena-infinity", "additional": {"app_name": "paypal-my-wallet", "difficulty": "hard", "verifier_path": "real-tasks/task_h25.py", "app_base_port": 8100}}}
{"query_id": "infinity-elation-clinical-records-task_h66", "dataset": "webarena-infinity", "query": "Create a new template called 'Anxiety Management' with HPI and Assessment sections, and billing code 99213 with description 'Office visit, established, low complexity'. Then create a visit note for Emily Nakamura using that new template and the Telehealth category, add a Psychological Status block to the note, and sign it.", "graders": ["infinity_state"], "start_url": "http://localhost:8000", "metadata": {"original_task_id": "elation-clinical-records-task_h66", "website": "elation-clinical-records", "category": "webarena-infinity", "additional": {"app_name": "elation-clinical-records", "difficulty": "hard", "verifier_path": "real-tasks/task_h66.py", "app_base_port": 8000}}}
{"query_id": "infinity-elation-clinical-records-task_h62", "dataset": "webarena-infinity", "query": "Look up which template is assigned to the COVID Vaccine appointment type. Remove all its existing document tags and replace them with the single tag 'COVID-Protocol'. Then also assign that same template to the Urgent Same-Day appointment type.", "graders": ["infinity_state"], "start_url": "http://localhost:8000", "metadata": {"original_task_id": "elation-clinical-records-task_h62", "website": "elation-clinical-records", "category": "webarena-infinity", "additional": {"app_name": "elation-clinical-records", "difficulty": "hard", "verifier_path": "real-tasks/task_h62.py", "app_base_port": 8000}}}
{"query_id": "infinity-elation-prescriptions-task_h32", "dataset": "webarena-infinity", "query": "The patient has a medication that's being dispensed as written (brand name only). Discontinue that prescription and replace it with a new one \u2014 same medication, same sig, same pharmacy \u2014 but allow generic substitution this time. Qty 30, 3 refills, 30 days supply.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h32", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h32.py", "app_base_port": 8020}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h48", "dataset": "webarena-infinity", "query": "Add the 'breaking-change' label to every open issue in the API v3 Migration epic and remove any existing workflow-scoped labels from those issues.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h48", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h48.py", "app_base_port": 8050}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h77", "dataset": "webarena-infinity", "query": "Rename the 'UX' label to 'user-experience', change its type to 'group', and then add it to every open issue in the Frontend Modernization epic that doesn't already have it.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h77", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h77.py", "app_base_port": 8050}}}
{"query_id": "infinity-xero-invoicing-task_h15", "dataset": "webarena-infinity", "query": "Create a new invoice for Summit Health Group for an annual software license and 12 months of support with a 10% discount on support.", "graders": ["infinity_state"], "start_url": "http://localhost:8120", "metadata": {"original_task_id": "xero-invoicing-task_h15", "website": "xero-invoicing", "category": "webarena-infinity", "additional": {"app_name": "xero-invoicing", "difficulty": "hard", "verifier_path": "real-tasks/task_h15.py", "app_base_port": 8120}}}
{"query_id": "infinity-elation-clinical-records-task_h55", "dataset": "webarena-infinity", "query": "Resolve every problem across all patients in the system that currently has a status of Controlled.", "graders": ["infinity_state"], "start_url": "http://localhost:8000", "metadata": {"original_task_id": "elation-clinical-records-task_h55", "website": "elation-clinical-records", "category": "webarena-infinity", "additional": {"app_name": "elation-clinical-records", "difficulty": "hard", "verifier_path": "real-tasks/task_h55.py", "app_base_port": 8000}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h8", "dataset": "webarena-infinity", "query": "Create a confidential issue titled 'Emergency security patch' with priority::critical and the 'security' label, assigned to James O'Brien and Oliver Schmidt, with weight 2 in the Security Hardening milestone.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h8", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h8.py", "app_base_port": 8050}}}
{"query_id": "infinity-paypal-my-wallet-task_h20", "dataset": "webarena-infinity", "query": "Make a $200 payment on PayPal Credit and change autopay to pay the full balance.", "graders": ["infinity_state"], "start_url": "http://localhost:8100", "metadata": {"original_task_id": "paypal-my-wallet-task_h20", "website": "paypal-my-wallet", "category": "webarena-infinity", "additional": {"app_name": "paypal-my-wallet", "difficulty": "hard", "verifier_path": "real-tasks/task_h20.py", "app_base_port": 8100}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h52", "dataset": "webarena-infinity", "query": "Create a new board called 'Performance Tracker' with lists for the priority::critical, priority::high, and priority::medium labels. Then add the 'priority::high' label to every open issue in the v4.1 milestone that has the 'performance' label.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h52", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h52.py", "app_base_port": 8050}}}
{"query_id": "infinity-paypal-my-wallet-task_h80", "dataset": "webarena-infinity", "query": "Save all available Food & Drink offers, buy a $25 DoorDash gift card for yourself, and switch currency conversion to use my card issuer.", "graders": ["infinity_state"], "start_url": "http://localhost:8100", "metadata": {"original_task_id": "paypal-my-wallet-task_h80", "website": "paypal-my-wallet", "category": "webarena-infinity", "additional": {"app_name": "paypal-my-wallet", "difficulty": "hard", "verifier_path": "real-tasks/task_h80.py", "app_base_port": 8100}}}
{"query_id": "infinity-gmail-accounts-and-contacts-task_h50", "dataset": "webarena-infinity", "query": "Add the Emergency label to every contact who is currently listed as a delegate (active, pending, or expired). Then remove all delegates whose status is not 'active'.", "graders": ["infinity_state"], "start_url": "http://localhost:8070", "metadata": {"original_task_id": "gmail-accounts-and-contacts-task_h50", "website": "gmail-accounts-and-contacts", "category": "webarena-infinity", "additional": {"app_name": "gmail-accounts-and-contacts", "difficulty": "hard", "verifier_path": "real-tasks/task_h50.py", "app_base_port": 8070}}}
{"query_id": "infinity-elation-clinical-records-task_h14", "dataset": "webarena-infinity", "query": "Add the tag 'Flu-Season' to every patient whose primary provider is Dr. Sarah Chen.", "graders": ["infinity_state"], "start_url": "http://localhost:8000", "metadata": {"original_task_id": "elation-clinical-records-task_h14", "website": "elation-clinical-records", "category": "webarena-infinity", "additional": {"app_name": "elation-clinical-records", "difficulty": "hard", "verifier_path": "real-tasks/task_h14.py", "app_base_port": 8000}}}
{"query_id": "infinity-figma-text-and-typography-task_h7", "dataset": "webarena-infinity", "query": "Remove all list formatting from every layer.", "graders": ["infinity_state"], "start_url": "http://localhost:8040", "metadata": {"original_task_id": "figma-text-and-typography-task_h7", "website": "figma-text-and-typography", "category": "webarena-infinity", "additional": {"app_name": "figma-text-and-typography", "difficulty": "hard", "verifier_path": "real-tasks/task_h7.py", "app_base_port": 8040}}}
{"query_id": "infinity-paypal-my-wallet-task_h26", "dataset": "webarena-infinity", "query": "Send a $50 Amazon gift card to sarah.chen@email.com with 'Thank you!' as the message, and save the Amazon cashback offer.", "graders": ["infinity_state"], "start_url": "http://localhost:8100", "metadata": {"original_task_id": "paypal-my-wallet-task_h26", "website": "paypal-my-wallet", "category": "webarena-infinity", "additional": {"app_name": "paypal-my-wallet", "difficulty": "hard", "verifier_path": "real-tasks/task_h26.py", "app_base_port": 8100}}}
{"query_id": "infinity-handshake-career-exploration-task_h97", "dataset": "webarena-infinity", "query": "Find the single most helpful answer across all Q&A questions and mark it helpful. Then find the most-viewed question and submit your own answer to it.", "graders": ["infinity_state"], "start_url": "http://localhost:8080", "metadata": {"original_task_id": "handshake-career-exploration-task_h97", "website": "handshake-career-exploration", "category": "webarena-infinity", "additional": {"app_name": "handshake-career-exploration", "difficulty": "hard", "verifier_path": "real-tasks/task_h97.py", "app_base_port": 8080}}}
{"query_id": "infinity-figma-slides-task_h79", "dataset": "webarena-infinity", "query": "In the adoption table, find the feature with the highest Target Q4 percentage. In the competitive table, change DesignCraft's entry for that same feature to 'Market Leader'. Then update that feature's Target Q4 to '95%'.", "graders": ["infinity_state"], "start_url": "http://localhost:8030", "metadata": {"original_task_id": "figma-slides-task_h79", "website": "figma-slides", "category": "webarena-infinity", "additional": {"app_name": "figma-slides", "difficulty": "hard", "verifier_path": "real-tasks/task_h79.py", "app_base_port": 8030}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h41", "dataset": "webarena-infinity", "query": "For every open issue in the v4.2 - Security Hardening milestone: if it is already confidential, set its health status to 'at risk'. If it is not confidential, make it confidential and set its health status to 'needs attention'.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h41", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h41.py", "app_base_port": 8050}}}
{"query_id": "infinity-handshake-career-exploration-task_h90", "dataset": "webarena-infinity", "query": "A student in the feed mentioned attending the NSBE conference. That student also answered a Q&A question about diversity programs in tech. Submit your own answer to that same question sharing your experience, then bookmark that student's feed post.", "graders": ["infinity_state"], "start_url": "http://localhost:8080", "metadata": {"original_task_id": "handshake-career-exploration-task_h90", "website": "handshake-career-exploration", "category": "webarena-infinity", "additional": {"app_name": "handshake-career-exploration", "difficulty": "hard", "verifier_path": "real-tasks/task_h90.py", "app_base_port": 8080}}}
{"query_id": "infinity-elation-prescriptions-task_h30", "dataset": "webarena-infinity", "query": "The patient has three temporary medications. Discontinue the corticosteroid taper and the penicillin antibiotic \u2014 the patient completed both courses. Move the remaining temporary medication to permanent Rx.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h30", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h30.py", "app_base_port": 8020}}}
{"query_id": "infinity-linear-account-settings-task_h19", "dataset": "webarena-infinity", "query": "Turn off all desktop application settings: open in desktop app, notification badge, and spell check.", "graders": ["infinity_state"], "start_url": "http://localhost:8090", "metadata": {"original_task_id": "linear-account-settings-task_h19", "website": "linear-account-settings", "category": "webarena-infinity", "additional": {"app_name": "linear-account-settings", "difficulty": "hard", "verifier_path": "real-tasks/task_h19.py", "app_base_port": 8090}}}
{"query_id": "infinity-elation-prescriptions-task_h39", "dataset": "webarena-infinity", "query": "Change the default pharmacy to Express Scripts Mail Pharmacy for mail-order prescriptions. Then document that the patient takes Magnesium Citrate 400mg tablet as an OTC supplement \u2014 once daily at bedtime, 30-day supply.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h39", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h39.py", "app_base_port": 8020}}}
{"query_id": "infinity-handshake-career-exploration-task_h136", "dataset": "webarena-infinity", "query": "Your earliest completed appointment was a specific type. Schedule a follow-up appointment of the same category and type with the same staff member, for March 28, 2026 at 9:00 AM, in person.", "graders": ["infinity_state"], "start_url": "http://localhost:8080", "metadata": {"original_task_id": "handshake-career-exploration-task_h136", "website": "handshake-career-exploration", "category": "webarena-infinity", "additional": {"app_name": "handshake-career-exploration", "difficulty": "hard", "verifier_path": "real-tasks/task_h136.py", "app_base_port": 8080}}}
{"query_id": "infinity-handshake-career-exploration-task_h105", "dataset": "webarena-infinity", "query": "Find the second-most-viewed question in Q&A. It has two answers \u2014 mark the one with fewer helpful votes as helpful.", "graders": ["infinity_state"], "start_url": "http://localhost:8080", "metadata": {"original_task_id": "handshake-career-exploration-task_h105", "website": "handshake-career-exploration", "category": "webarena-infinity", "additional": {"app_name": "handshake-career-exploration", "difficulty": "hard", "verifier_path": "real-tasks/task_h105.py", "app_base_port": 8080}}}
{"query_id": "infinity-gmail-accounts-and-contacts-task_h22", "dataset": "webarena-infinity", "query": "The Engineering Manager at TechCorp is listed as one of your delegates. Remove her delegation and unstar her contact.", "graders": ["infinity_state"], "start_url": "http://localhost:8070", "metadata": {"original_task_id": "gmail-accounts-and-contacts-task_h22", "website": "gmail-accounts-and-contacts", "category": "webarena-infinity", "additional": {"app_name": "gmail-accounts-and-contacts", "difficulty": "hard", "verifier_path": "real-tasks/task_h22.py", "app_base_port": 8070}}}
{"query_id": "infinity-elation-patient-communication-task_h9", "dataset": "webarena-infinity", "query": "Acknowledge all unacknowledged reminders in the system.", "graders": ["infinity_state"], "start_url": "http://localhost:8010", "metadata": {"original_task_id": "elation-patient-communication-task_h9", "website": "elation-patient-communication", "category": "webarena-infinity", "additional": {"app_name": "elation-patient-communication", "difficulty": "hard", "verifier_path": "real-tasks/task_h9.py", "app_base_port": 8010}}}
{"query_id": "infinity-superhuman-general-task_h1", "dataset": "webarena-infinity", "query": "Label the FinancePlus partnership email and the QuantumLab prototype email as 'Clients'.", "graders": ["infinity_state"], "start_url": "http://localhost:8110", "metadata": {"original_task_id": "superhuman-general-task_h1", "website": "superhuman-general", "category": "webarena-infinity", "additional": {"app_name": "superhuman-general", "difficulty": "hard", "verifier_path": "real-tasks/task_h1.py", "app_base_port": 8110}}}
{"query_id": "infinity-xero-invoicing-task_h79", "dataset": "webarena-infinity", "query": "Change the invoice prefix to 'AUS-' and the next number to 100, then create a new invoice for CloudNine Analytics for 8 hours of UI/UX design work.", "graders": ["infinity_state"], "start_url": "http://localhost:8120", "metadata": {"original_task_id": "xero-invoicing-task_h79", "website": "xero-invoicing", "category": "webarena-infinity", "additional": {"app_name": "xero-invoicing", "difficulty": "hard", "verifier_path": "real-tasks/task_h79.py", "app_base_port": 8120}}}
{"query_id": "infinity-figma-slides-task_h16", "dataset": "webarena-infinity", "query": "Enable slide numbers on every slide using the 'with total' format and change the aspect ratio to 4:3.", "graders": ["infinity_state"], "start_url": "http://localhost:8030", "metadata": {"original_task_id": "figma-slides-task_h16", "website": "figma-slides", "category": "webarena-infinity", "additional": {"app_name": "figma-slides", "difficulty": "hard", "verifier_path": "real-tasks/task_h16.py", "app_base_port": 8030}}}
{"query_id": "infinity-linear-account-settings-task_h16", "dataset": "webarena-infinity", "query": "Revoke all API keys that have an expiration date.", "graders": ["infinity_state"], "start_url": "http://localhost:8090", "metadata": {"original_task_id": "linear-account-settings-task_h16", "website": "linear-account-settings", "category": "webarena-infinity", "additional": {"app_name": "linear-account-settings", "difficulty": "hard", "verifier_path": "real-tasks/task_h16.py", "app_base_port": 8090}}}
{"query_id": "infinity-elation-prescriptions-task_h2", "dataset": "webarena-infinity", "query": "Prescribe Buspirone 10mg for the patient's anxiety \u2014 once daily in the morning, qty 30, 5 refills. Send it to the same pharmacy that fills his Sertraline.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h2", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h2.py", "app_base_port": 8020}}}
{"query_id": "infinity-handshake-career-exploration-task_h1", "dataset": "webarena-infinity", "query": "Follow all consulting firms on Handshake.", "graders": ["infinity_state"], "start_url": "http://localhost:8080", "metadata": {"original_task_id": "handshake-career-exploration-task_h1", "website": "handshake-career-exploration", "category": "webarena-infinity", "additional": {"app_name": "handshake-career-exploration", "difficulty": "hard", "verifier_path": "real-tasks/task_h1.py", "app_base_port": 8080}}}
{"query_id": "infinity-handshake-career-exploration-task_h141", "dataset": "webarena-infinity", "query": "Some of your saved jobs are from employers you haven't followed yet. Find and follow each of those employers.", "graders": ["infinity_state"], "start_url": "http://localhost:8080", "metadata": {"original_task_id": "handshake-career-exploration-task_h141", "website": "handshake-career-exploration", "category": "webarena-infinity", "additional": {"app_name": "handshake-career-exploration", "difficulty": "hard", "verifier_path": "real-tasks/task_h141.py", "app_base_port": 8080}}}
{"query_id": "infinity-figma-text-and-typography-task_h74", "dataset": "webarena-infinity", "query": "Set the spelling language to Japanese, the big nudge amount to 50, and the default horizontal alignment to right.", "graders": ["infinity_state"], "start_url": "http://localhost:8040", "metadata": {"original_task_id": "figma-text-and-typography-task_h74", "website": "figma-text-and-typography", "category": "webarena-infinity", "additional": {"app_name": "figma-text-and-typography", "difficulty": "hard", "verifier_path": "real-tasks/task_h74.py", "app_base_port": 8040}}}
{"query_id": "infinity-elation-patient-communication-task_h63", "dataset": "webarena-infinity", "query": "Check the visit summaries to find the patient whose BNP level improved. Reply to their most recent message confirming they can resume light activity, then update their emergency contact's phone number to (650) 555-0001.", "graders": ["infinity_state"], "start_url": "http://localhost:8010", "metadata": {"original_task_id": "elation-patient-communication-task_h63", "website": "elation-patient-communication", "category": "webarena-infinity", "additional": {"app_name": "elation-patient-communication", "difficulty": "hard", "verifier_path": "real-tasks/task_h63.py", "app_base_port": 8010}}}
{"query_id": "infinity-elation-patient-communication-task_h14", "dataset": "webarena-infinity", "query": "Change Dr. Torres's notification timeframe to 'Do not notify me' and remove Dr. Torres from Dr. Chen's General Question routing.", "graders": ["infinity_state"], "start_url": "http://localhost:8010", "metadata": {"original_task_id": "elation-patient-communication-task_h14", "website": "elation-patient-communication", "category": "webarena-infinity", "additional": {"app_name": "elation-patient-communication", "difficulty": "hard", "verifier_path": "real-tasks/task_h14.py", "app_base_port": 8010}}}
{"query_id": "infinity-gitlab-plan-and-track-task_h67", "dataset": "webarena-infinity", "query": "Delete all time entries from the GraphQL gateway issue, add a single new entry of 16 hours with summary 'Complete rewrite estimate', and set its time estimate to 40 hours.", "graders": ["infinity_state"], "start_url": "http://localhost:8050", "metadata": {"original_task_id": "gitlab-plan-and-track-task_h67", "website": "gitlab-plan-and-track", "category": "webarena-infinity", "additional": {"app_name": "gitlab-plan-and-track", "difficulty": "hard", "verifier_path": "real-tasks/task_h67.py", "app_base_port": 8050}}}
{"query_id": "infinity-gmail-accounts-and-contacts-task_h73", "dataset": "webarena-infinity", "query": "Among the individual people in your other contacts (those with a first and last name), find the one who was saved most recently. Move them to your main contacts, set their company to 'Salesforce', job title to 'Account Executive', and add the Work label.", "graders": ["infinity_state"], "start_url": "http://localhost:8070", "metadata": {"original_task_id": "gmail-accounts-and-contacts-task_h73", "website": "gmail-accounts-and-contacts", "category": "webarena-infinity", "additional": {"app_name": "gmail-accounts-and-contacts", "difficulty": "hard", "verifier_path": "real-tasks/task_h73.py", "app_base_port": 8070}}}
{"query_id": "infinity-elation-prescriptions-task_h4", "dataset": "webarena-infinity", "query": "Run a medication reconciliation and mark the Calcium+D3 supplement for discontinuation during the review.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h4", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h4.py", "app_base_port": 8020}}}
{"query_id": "infinity-elation-prescriptions-task_h47", "dataset": "webarena-infinity", "query": "The patient's SSRI is currently dispensed at a different pharmacy than most of his other medications. Prescribe a refill of the same SSRI at the same dose and sig, but send it to CVS #4521 instead \u2014 qty 30, 5 refills, 30 days supply.", "graders": ["infinity_state"], "start_url": "http://localhost:8020", "metadata": {"original_task_id": "elation-prescriptions-task_h47", "website": "elation-prescriptions", "category": "webarena-infinity", "additional": {"app_name": "elation-prescriptions", "difficulty": "hard", "verifier_path": "real-tasks/task_h47.py", "app_base_port": 8020}}}
{"query_id": "infinity-paypal-my-wallet-task_h89", "dataset": "webarena-infinity", "query": "If your USD PayPal balance is above $2,500, convert $500 to Japanese Yen. If it is $2,500 or below, first add $500 from your Chase bank account, then convert $500 to JPY. Either way, set the debit card cash back category to Fuel.", "graders": ["infinity_state"], "start_url": "http://localhost:8100", "metadata": {"original_task_id": "paypal-my-wallet-task_h89", "website": "paypal-my-wallet", "category": "webarena-infinity", "additional": {"app_name": "paypal-my-wallet", "difficulty": "hard", "verifier_path": "real-tasks/task_h89.py", "app_base_port": 8100}}}

View File

@@ -0,0 +1,88 @@
#!/usr/bin/env python3
"""
AGI SDK evaluation helper for BrowserOS eval framework.
Reads JSON from stdin with task_id and env_state, runs the agisdk
evaluator, and outputs the result as JSON to stdout.
Input format:
{"task_id": "dashdish-1", "env_state": {...}, "model_response": ""}
Output format:
{"reward": 0.0, "pass": false, "message": "...", "per_criterion": [...]}
"""
import json
import sys
def main():
data = json.loads(sys.stdin.read())
task_id = data["task_id"]
env_state = data["env_state"]
model_response = data.get("model_response", "")
try:
from agisdk.REAL.browsergym.webclones.evaluate import WebCloneEvaluator
from agisdk.REAL.browsergym.webclones.task_config import TaskConfig
except ImportError:
print(
json.dumps(
{
"reward": 0,
"pass": False,
"message": "agisdk package not installed. Run: pip install agisdk",
"per_criterion": [],
}
)
)
sys.exit(0)
try:
# Redirect stdout to stderr during evaluation — agisdk's rich logger
# prints directly to stdout, which would corrupt our JSON output
real_stdout = sys.stdout
sys.stdout = sys.stderr
tc = TaskConfig(task_id)
evaluator = WebCloneEvaluator(tc)
reward_val, _done, message, info = evaluator.evaluate(
env_state=env_state, model_response=model_response
)
sys.stdout = real_stdout
reward_val = float(reward_val) if reward_val is not None else 0.0
results = info.get("results", [])
per_criterion = [
{"passed": r[0], "detail": str(r[1]) if len(r) > 1 else ""}
for r in results
]
print(
json.dumps(
{
"reward": reward_val,
"pass": reward_val == 1.0,
"message": str(message),
"per_criterion": per_criterion,
}
)
)
except Exception as e:
sys.stdout = real_stdout if "real_stdout" in dir() else sys.__stdout__
print(
json.dumps(
{
"reward": 0,
"pass": False,
"message": f"Evaluation error: {str(e)}",
"per_criterion": [],
}
)
)
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,92 @@
#!/usr/bin/env python3
"""
Build JSONL dataset for AGI SDK / REAL Bench evaluation.
Reads task definitions from the agisdk package, filters to feasible
action-only tasks (excludes llm_boolean evaluators), and outputs JSONL
to stdout in the BrowserOS eval framework format.
Usage:
python scripts/build-agisdk-dataset.py > data/agisdk-real.jsonl
"""
import json
import sys
# evals-omnizon.vercel.app was DMCA-takedown'd by Vercel (HTTP 451). Every task
# on that site fails grading with "Failed to fetch /finish endpoint".
EXCLUDED_WEBSITES = {"omnizon"}
def has_llm_eval(task: dict) -> bool:
return any(e.get("type") == "llm_boolean" for e in task.get("evals", []))
def main():
try:
from agisdk.REAL.tasks import all_tasks
except ImportError:
print(
"Error: agisdk package not installed. Run: pip install agisdk",
file=sys.stderr,
)
sys.exit(1)
count = 0
skipped_infeasible = 0
skipped_llm = 0
skipped_excluded = 0
for task in all_tasks:
if not task.get("possible", True):
skipped_infeasible += 1
continue
if has_llm_eval(task):
skipped_llm += 1
continue
website = task.get("website", {})
if website.get("id") in EXCLUDED_WEBSITES:
skipped_excluded += 1
continue
task_id = task["id"]
goal = task.get("goal", "")
start_url = website.get("url", "")
if not start_url or not goal:
print(f"Warning: Skipping {task_id} — missing url or goal", file=sys.stderr)
continue
entry = {
"query_id": f"agisdk-{task_id}",
"dataset": "agisdk-real",
"query": goal,
"graders": ["agisdk_state_diff"],
"start_url": start_url,
"metadata": {
"original_task_id": task_id,
"website": website.get("name", ""),
"category": "agisdk-real",
"additional": {
"agisdk_task_id": task_id,
"challenge_type": task.get("challengeType", "action"),
"difficulty": task.get("difficulty", "unknown"),
"similar_to": website.get("similarTo", ""),
},
},
}
print(json.dumps(entry))
count += 1
print(
f"Generated {count} tasks (skipped {skipped_infeasible} infeasible, "
f"{skipped_llm} llm_boolean, {skipped_excluded} excluded sites)",
file=sys.stderr,
)
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,118 @@
#!/usr/bin/env python3
"""
Dataset generator for WebArena-Infinity benchmark.
Reads real-tasks.json from each app directory and outputs JSONL
in the eval framework's TaskSchema format.
Usage:
python build-infinity-dataset.py --apps-dir /path/to/webarena-infinity/apps
python build-infinity-dataset.py --apps-dir /path/to/apps --apps gmail linear --difficulty medium
"""
import argparse
import json
import os
import sys
def load_tasks(app_dir: str) -> list[dict]:
tasks_file = os.path.join(app_dir, "real-tasks.json")
if not os.path.exists(tasks_file):
print(f"Warning: No real-tasks.json found in {app_dir}", file=sys.stderr)
return []
with open(tasks_file) as f:
return json.load(f)
def build_task_entry(
app_name: str,
task: dict,
base_port: int,
) -> dict:
task_id = task.get("id", task.get("task_id", "unknown"))
difficulty = task.get("difficulty", "unknown")
query = task.get("query", task.get("instruction", task.get("task", "")))
verifier_path = task.get(
"verify",
task.get("verifier_path", f"real-tasks/{task_id}.py"),
)
return {
"query_id": f"infinity-{app_name}-{task_id}",
"dataset": "webarena-infinity",
"query": query,
"graders": ["infinity_state"],
"start_url": f"http://localhost:{base_port}",
"setup_script": f"POST http://localhost:{base_port}/api/reset",
"metadata": {
"original_task_id": f"{app_name}-{task_id}",
"website": app_name,
"category": "webarena-infinity",
"additional": {
"app_name": app_name,
"difficulty": difficulty,
"verifier_path": verifier_path,
"app_base_port": base_port,
},
},
}
def main():
parser = argparse.ArgumentParser(
description="Generate JSONL dataset from WebArena-Infinity apps"
)
parser.add_argument(
"--apps-dir",
required=True,
help="Path to webarena-infinity/apps/ directory",
)
parser.add_argument(
"--apps",
nargs="*",
default=None,
help="Filter to specific app names (default: all)",
)
parser.add_argument(
"--difficulty",
choices=["easy", "medium", "hard"],
default=None,
help="Filter by difficulty tier",
)
parser.add_argument(
"--base-port",
type=int,
default=8000,
help="Starting port number for apps (default: 8000)",
)
args = parser.parse_args()
if not os.path.isdir(args.apps_dir):
print(f"Error: {args.apps_dir} is not a directory", file=sys.stderr)
sys.exit(1)
app_dirs = sorted(os.listdir(args.apps_dir))
if args.apps:
app_dirs = [d for d in app_dirs if d in args.apps]
port = args.base_port
for app_name in app_dirs:
app_path = os.path.join(args.apps_dir, app_name)
if not os.path.isdir(app_path):
continue
tasks = load_tasks(app_path)
for task in tasks:
difficulty = task.get("difficulty", "unknown")
if args.difficulty and difficulty != args.difficulty:
continue
entry = build_task_entry(app_name, task, port)
print(json.dumps(entry))
port += 1
if __name__ == "__main__":
main()

View File

@@ -0,0 +1,82 @@
#!/usr/bin/env python3
"""
Evaluation helper for WebArena-Infinity verifier scripts.
Reads JSON from stdin with app_server_url, verifier_path, and task_id.
Runs the verifier against the app server and outputs a JSON result.
Verifiers have the signature: verify(server_url: str) -> tuple[bool, str]
They fetch /api/state internally and return (passed, message).
Usage:
echo '{"app_server_url": "http://localhost:8000", "verifier_path": "/path/to/verify.py"}' | python infinity-evaluate.py
"""
import importlib.util
import json
import sys
import traceback
def load_verifier(verifier_path: str):
spec = importlib.util.spec_from_file_location("verifier", verifier_path)
if spec is None or spec.loader is None:
raise ImportError(f"Cannot load verifier from {verifier_path}")
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
def main():
try:
data = json.loads(sys.stdin.read())
except json.JSONDecodeError as e:
print(json.dumps({"pass": False, "reward": 0.0, "message": f"Invalid JSON input: {e}"}))
sys.exit(1)
server_url = data.get("app_server_url", "")
verifier_path = data.get("verifier_path", "")
if not server_url or not verifier_path:
print(json.dumps({
"pass": False,
"reward": 0.0,
"message": "Missing app_server_url or verifier_path",
}))
sys.exit(1)
try:
verifier = load_verifier(verifier_path)
fn = getattr(verifier, "verify", None)
if not callable(fn):
raise AttributeError(
f"Verifier has no verify() function. "
f"Available: {[a for a in dir(verifier) if not a.startswith('_')]}"
)
# Verifiers take server_url and fetch state internally
result = fn(server_url)
# Return is tuple[bool, str]
if isinstance(result, tuple) and len(result) >= 2:
passed, message = result[0], str(result[1])
else:
passed, message = bool(result), str(result)
except Exception as e:
print(json.dumps({
"pass": False,
"reward": 0.0,
"message": f"Verifier error: {e}\n{traceback.format_exc()}",
}))
sys.exit(1)
print(json.dumps({
"pass": passed,
"reward": 1.0 if passed else 0.0,
"message": message,
}))
if __name__ == "__main__":
main()

View File

@@ -59,6 +59,8 @@ interface RunSummary {
}
const PASS_FAIL_GRADER_ORDER = [
'agisdk_state_diff',
'infinity_state',
'performance_grader',
'webvoyager_grader',
'fara_combined',

View File

@@ -0,0 +1,202 @@
import { spawn } from 'node:child_process'
import { join } from 'node:path'
import type { GraderResult } from '../../types'
import { callMcpTool } from '../../utils/mcp-client'
import type { Grader, GraderInput } from '../types'
const EVAL_SCRIPT = join(
import.meta.dirname,
'..',
'..',
'..',
'scripts',
'agisdk-evaluate.py',
)
export class AgisdkStateDiffGrader implements Grader {
name = 'agisdk_state_diff'
async grade(input: GraderInput): Promise<GraderResult> {
const taskId = this.extractTaskId(input.task.query_id)
const startUrl = this.extractStartUrl(input)
const mcpEndpoint =
input.mcpUrl ||
`${process.env.BROWSEROS_SERVER_URL || 'http://127.0.0.1:9110'}/mcp`
if (!startUrl) {
return {
score: 0,
pass: false,
reasoning: 'Could not determine clone site URL from task',
}
}
const origin = new URL(startUrl).origin
let envState: Record<string, unknown>
try {
envState = await this.fetchFinishState(origin, mcpEndpoint)
} catch (error) {
return {
score: 0,
pass: false,
reasoning: `Failed to fetch /finish endpoint: ${error instanceof Error ? error.message : String(error)}`,
details: { origin, error: true },
}
}
try {
const result = await this.runPythonEvaluator(
taskId,
envState,
input.finalAnswer || '',
)
return {
score: result.reward,
pass: result.pass,
reasoning:
result.message ||
(result.pass ? 'All criteria passed' : 'Some criteria failed'),
details: {
reward: result.reward,
per_criterion: result.per_criterion,
origin,
agisdk_task_id: taskId,
},
}
} catch (error) {
return {
score: 0,
pass: false,
reasoning: `Python evaluator error: ${error instanceof Error ? error.message : String(error)}`,
details: { error: true },
}
}
}
private extractTaskId(queryId: string): string {
return queryId.replace(/^agisdk-/, '')
}
private extractStartUrl(input: GraderInput): string | null {
// Derive from task_id: "dashdish-10" → "https://evals-dashdish.vercel.app"
// Task IDs are "{site}-{number}" where site may contain hyphens (e.g. "fly-unified-5")
const taskId = this.extractTaskId(input.task.query_id)
const siteId = taskId.replace(/-\d+$/, '')
if (siteId) return `https://evals-${siteId}.vercel.app`
// Fallback: search messages for vercel.app URLs
for (const msg of input.messages) {
const text =
msg.type === 'user'
? msg.content
: msg.type === 'tool-input-available'
? JSON.stringify(msg.input)
: ''
const urlMatch = text.match(/https?:\/\/[^\s"']+\.vercel\.app/)
if (urlMatch) return urlMatch[0]
}
return null
}
private async fetchFinishState(
origin: string,
mcpEndpoint: string,
): Promise<Record<string, unknown>> {
const finishUrl = `${origin}/finish`
// Navigate browser to /finish page (state diff is rendered client-side)
await callMcpTool(mcpEndpoint, 'navigate_page', {
url: finishUrl,
page: 1,
})
// Wait for the page to render, then extract JSON from <pre> element
const result = await callMcpTool(mcpEndpoint, 'evaluate_script', {
page: 1,
expression: `
new Promise((resolve, reject) => {
let attempts = 0;
const check = () => {
const pre = document.querySelector('pre');
if (pre && pre.textContent.trim().startsWith('{')) {
resolve(pre.textContent);
} else if (++attempts > 20) {
reject(new Error('Timed out waiting for <pre> JSON on /finish'));
} else {
setTimeout(check, 500);
}
};
check();
})
`,
})
const textContent = result.content?.find(
(c: { type: string }) => c.type === 'text',
)
if (!textContent?.text) {
throw new Error('No text content returned from /finish page')
}
return JSON.parse(textContent.text) as Record<string, unknown>
}
private runPythonEvaluator(
taskId: string,
envState: Record<string, unknown>,
modelResponse: string,
): Promise<{
reward: number
pass: boolean
message: string
per_criterion: unknown[]
}> {
return new Promise((resolve, reject) => {
const proc = spawn('python3', [EVAL_SCRIPT], {
stdio: ['pipe', 'pipe', 'pipe'],
})
const inputData = JSON.stringify({
task_id: taskId,
env_state: envState,
model_response: modelResponse,
})
let stdout = ''
let stderr = ''
proc.stdout.on('data', (data: Buffer) => {
stdout += data.toString()
})
proc.stderr.on('data', (data: Buffer) => {
stderr += data.toString()
})
proc.on('close', (code) => {
if (code !== 0) {
reject(
new Error(`Python evaluator exited with code ${code}: ${stderr}`),
)
return
}
try {
const result = JSON.parse(stdout.trim())
resolve(result)
} catch {
reject(new Error(`Failed to parse evaluator output: ${stdout}`))
}
})
proc.on('error', (err) => {
reject(new Error(`Failed to spawn Python evaluator: ${err.message}`))
})
proc.stdin.write(inputData)
proc.stdin.end()
})
}
}

View File

@@ -0,0 +1,134 @@
import { join, resolve } from 'node:path'
import type { GraderResult } from '../../types'
import type { Grader, GraderInput } from '../types'
interface InfinityEvalInput {
app_server_url: string
verifier_path: string
task_id: string
}
interface InfinityEvalOutput {
pass: boolean
reward: number
message: string
}
const EVAL_SCRIPT = resolve(
import.meta.dir,
'../../../scripts/infinity-evaluate.py',
)
export class InfinityStateGrader implements Grader {
name = 'infinity_state'
async grade(input: GraderInput): Promise<GraderResult> {
const parsed = this.parseQueryId(input.task.query_id)
if (!parsed) {
return {
score: 0,
pass: false,
reasoning: `Cannot parse query_id "${input.task.query_id}" — expected format: infinity-{app}-{task_id}`,
}
}
const appServerUrl = this.resolveAppServerUrl(input)
if (!appServerUrl) {
return {
score: 0,
pass: false,
reasoning: 'Cannot determine app server URL',
}
}
const infinityDir = process.env.WEBARENA_INFINITY_DIR
if (!infinityDir) {
return {
score: 0,
pass: false,
reasoning:
'WEBARENA_INFINITY_DIR env var not set. Point it to the webarena-infinity repo root.',
}
}
const verifierPath = join(
infinityDir,
'apps',
parsed.appName,
'real-tasks',
`${parsed.taskId}.py`,
)
const evalInput: InfinityEvalInput = {
app_server_url: appServerUrl,
verifier_path: verifierPath,
task_id: input.task.query_id,
}
try {
const result = await this.runPythonEvaluator(evalInput)
return {
score: result.pass ? 1 : 0,
pass: result.pass,
reasoning: result.message,
details: {
reward: result.reward,
app_name: parsed.appName,
app_server_url: appServerUrl,
},
}
} catch (error) {
return {
score: 0,
pass: false,
reasoning: `Evaluator process error: ${error instanceof Error ? error.message : String(error)}`,
}
}
}
private parseQueryId(
queryId: string,
): { appName: string; taskId: string } | null {
// Task IDs start with "task_", app names may contain hyphens
// e.g. "infinity-elation-prescriptions-task_h69"
const match = queryId.match(/^infinity-(.+)-(task_.+)$/)
if (!match) return null
return { appName: match[1], taskId: match[2] }
}
private resolveAppServerUrl(input: GraderInput): string | null {
// Passed directly from task executor (started by InfinityAppManager)
if (input.infinityAppUrl) return input.infinityAppUrl
// Fallback: env var for manual testing
if (process.env.INFINITY_APP_URL) return process.env.INFINITY_APP_URL
return null
}
private async runPythonEvaluator(
evalInput: InfinityEvalInput,
): Promise<InfinityEvalOutput> {
const proc = Bun.spawn(['python3', EVAL_SCRIPT], {
stdin: 'pipe',
stdout: 'pipe',
stderr: 'pipe',
})
const inputJson = JSON.stringify(evalInput)
proc.stdin.write(inputJson)
proc.stdin.end()
const stdout = await new Response(proc.stdout).text()
const stderr = await new Response(proc.stderr).text()
const exitCode = await proc.exited
if (exitCode !== 0) {
throw new Error(
`Python evaluator exited with code ${exitCode}: ${stderr || stdout}`,
)
}
return JSON.parse(stdout.trim()) as InfinityEvalOutput
}
}

View File

@@ -1,4 +1,6 @@
import type { GraderResult } from '../types'
import { AgisdkStateDiffGrader } from './benchmark/agisdk-state-diff'
import { InfinityStateGrader } from './benchmark/infinity-state'
import { Mind2WebJudgeGrader } from './benchmark/mind2web'
import { WebVoyagerGrader } from './benchmark/webvoyager'
import { FaraAlignmentGrader } from './fara/alignment'
@@ -19,7 +21,13 @@ export function createGrader(
options: GraderOptions | null,
): Grader | null {
switch (name) {
// Benchmark graders
// Deterministic benchmark graders (no LLM judge)
case 'agisdk_state_diff':
return new AgisdkStateDiffGrader()
case 'infinity_state':
return new InfinityStateGrader()
// LLM-based benchmark graders
case 'webvoyager_grader':
if (!options?.apiKey) return null
return new WebVoyagerGrader(
@@ -107,10 +115,12 @@ export async function runGraders(
// Export grader classes for direct use
export {
AgisdkStateDiffGrader,
FaraAlignmentGrader,
FaraCombinedGrader,
FaraMultimodalGrader,
FaraRubricGrader,
InfinityStateGrader,
Mind2WebJudgeGrader,
PerformanceGrader,
WebVoyagerGrader,

View File

@@ -11,6 +11,8 @@ export interface GraderInput {
finalAnswer: string | null
expectedAnswer?: string | null
outputDir: string
mcpUrl?: string
infinityAppUrl?: string
}
export interface Grader {

View File

@@ -0,0 +1,89 @@
/**
* Manages WebArena-Infinity app server lifecycle per task.
*
* Each worker gets a unique port: base_port + worker_index.
* Server is started fresh before each task and killed after,
* guaranteeing clean state.
*/
import { type ChildProcess, spawn } from 'node:child_process'
import { join } from 'node:path'
export class InfinityAppManager {
private proc: ChildProcess | null = null
private port: number
private infinityDir: string
constructor(
private workerIndex: number,
private basePort: number = 8000,
) {
this.port = basePort + workerIndex
this.infinityDir = process.env.WEBARENA_INFINITY_DIR || ''
}
async startApp(appName: string): Promise<string> {
await this.stop()
if (!this.infinityDir) {
throw new Error('WEBARENA_INFINITY_DIR env var not set')
}
const serverScript = join(this.infinityDir, 'apps', appName, 'server.py')
this.proc = spawn('python3', [serverScript, '--port', String(this.port)], {
stdio: ['ignore', 'pipe', 'pipe'],
cwd: join(this.infinityDir, 'apps', appName),
})
// Wait for server to be ready
const url = `http://localhost:${this.port}`
await this.waitForReady(url)
return url
}
async stop(): Promise<void> {
if (this.proc) {
this.proc.kill('SIGTERM')
await new Promise<void>((resolve) => {
const timeout = setTimeout(() => {
this.proc?.kill('SIGKILL')
resolve()
}, 3000)
this.proc?.on('exit', () => {
clearTimeout(timeout)
resolve()
})
})
this.proc = null
}
}
getPort(): number {
return this.port
}
getUrl(): string {
return `http://localhost:${this.port}`
}
private async waitForReady(
url: string,
maxAttempts = 30,
intervalMs = 500,
): Promise<void> {
for (let i = 0; i < maxAttempts; i++) {
try {
const resp = await fetch(url, {
signal: AbortSignal.timeout(2000),
})
if (resp.ok) return
} catch {
// Server not ready yet
}
await new Promise((r) => setTimeout(r, intervalMs))
}
throw new Error(
`Infinity app server not ready after ${maxAttempts * intervalMs}ms on port ${this.port}`,
)
}
}

View File

@@ -160,6 +160,7 @@ export class ParallelExecutor {
}
const executor = createTaskExecutor(
workerConfig,
workerIndex,
this.config.outputDir,
this.config.graderOptions,
this.config.onEvent,

View File

@@ -9,6 +9,7 @@ import {
import { runGraders } from '../graders/registry'
import type { ErrorSource, EvalConfig, GraderResult, Task } from '../types'
import { callMcpTool } from '../utils/mcp-client'
import { InfinityAppManager } from './infinity-app-manager'
import type { GraderOptions, TaskResult } from './types'
// ============================================================================
@@ -46,6 +47,7 @@ export interface TaskExecutorDeps {
export class TaskExecutor {
constructor(
private readonly config: EvalConfig,
private readonly workerIndex: number,
private readonly outputDir: string,
private readonly deps: TaskExecutorDeps,
) {}
@@ -101,6 +103,35 @@ export class TaskExecutor {
// Resolve page ID once — fresh browser has exactly one page
const pageId = await this.resolveInitialPageId(mcpUrl)
// For Infinity tasks, start a fresh app server per task
let infinityManager: InfinityAppManager | null = null
let actualStartUrl = task.start_url
if (task.dataset === 'webarena-infinity') {
const appName = (task.metadata?.additional as Record<string, unknown>)
?.app_name as string
const appBasePort =
((task.metadata?.additional as Record<string, unknown>)
?.app_base_port as number) || 8000
if (appName && process.env.WEBARENA_INFINITY_DIR) {
infinityManager = new InfinityAppManager(this.workerIndex, appBasePort)
try {
actualStartUrl = await infinityManager.startApp(appName)
console.log(
` Infinity app "${appName}" started on port ${infinityManager.getPort()}`,
)
} catch (error) {
throw new TaskExecutionError(
`Failed to start Infinity app: ${error instanceof Error ? error.message : String(error)}`,
task,
'navigation',
error instanceof Error ? error : undefined,
)
}
}
}
try {
// Phase 1: Set viewport + navigate to start URL
try {
@@ -114,10 +145,10 @@ export class TaskExecutor {
)
}
if (task.start_url && task.start_url !== 'about:blank') {
if (actualStartUrl && actualStartUrl !== 'about:blank') {
try {
await callMcpTool(mcpUrl, 'navigate_page', {
url: task.start_url,
url: actualStartUrl,
page: pageId,
})
} catch (error) {
@@ -134,7 +165,11 @@ export class TaskExecutor {
const agentResult = await this.executeAgent(task, pageId)
// Phase 3: Run graders
const graderResults = await this.runGraders(task, agentResult)
const graderResults = await this.runGraders(
task,
agentResult,
infinityManager?.getUrl(),
)
const status =
agentResult.metadata.termination_reason === 'timeout'
@@ -169,6 +204,11 @@ export class TaskExecutor {
} catch {
// Ignore cleanup errors
}
// Stop Infinity app server if running
if (infinityManager) {
await infinityManager.stop().catch(() => {})
}
}
}
@@ -209,6 +249,7 @@ export class TaskExecutor {
private async runGraders(
task: Task,
agentResult: AgentResult,
infinityAppUrl?: string,
): Promise<Record<string, GraderResult>> {
const configGraders = this.config.graders ?? []
const taskGraders = task.graders ?? []
@@ -234,6 +275,8 @@ export class TaskExecutor {
expectedAnswer: (task.metadata?.additional as Record<string, unknown>)
?.answer as string | undefined,
outputDir: join(this.outputDir, task.query_id),
mcpUrl: `${this.config.browseros.server_url}/mcp`,
infinityAppUrl,
},
this.deps.graderOptions,
)
@@ -269,11 +312,12 @@ export class TaskExecutor {
export function createTaskExecutor(
config: EvalConfig,
workerIndex: number,
outputDir: string,
graderOptions: GraderOptions | null,
onEvent?: (taskId: string, event: Record<string, unknown>) => void,
): TaskExecutor {
return new TaskExecutor(config, outputDir, {
return new TaskExecutor(config, workerIndex, outputDir, {
graderOptions,
onEvent,
})

View File

@@ -100,6 +100,8 @@ export interface TaskResultSummary {
// ============================================================================
export const PASS_FAIL_GRADER_ORDER = [
'agisdk_state_diff',
'infinity_state',
'performance_grader',
'webvoyager_grader',
'fara_combined',

View File

@@ -7,6 +7,11 @@ BROWSEROS_EXTENSION_PORT=9300
# BROWSEROS_RESOURCES_DIR=./resources
# BROWSEROS_EXECUTION_DIR=./out
# VM cache (optional - runtime downloads published agent cache in background)
# Set prefetch=false to skip startup warmup; VM/OpenClaw startup still syncs on demand.
BROWSEROS_VM_CACHE_PREFETCH=true
BROWSEROS_VM_CACHE_MANIFEST_URL=https://cdn.browseros.com/vm/manifest.json
# BrowserOS config
BROWSEROS_CONFIG_URL=https://llm.browseros.com/api/browseros-server/config
BROWSEROS_VERSION=

View File

@@ -5,6 +5,9 @@ CODEGEN_SERVICE_URL=
POSTHOG_API_KEY=
SENTRY_DSN=
BROWSEROS_VM_CACHE_PREFETCH=true
BROWSEROS_VM_CACHE_MANIFEST_URL=https://cdn.browseros.com/vm/manifest.json
R2_ACCOUNT_ID=
R2_ACCESS_KEY_ID=
R2_SECRET_ACCESS_KEY=

View File

@@ -142,7 +142,7 @@ cp .env.example .env.development
bun run start
```
See the [agent monorepo README](../../README.md) for full environment variable reference and `process-compose` setup.
See the [agent monorepo README](../../README.md) for full environment variable reference and `dev:watch` setup.
### Testing

View File

@@ -1,6 +1,6 @@
{
"name": "@browseros/server",
"version": "0.0.85",
"version": "0.0.92",
"description": "BrowserOS server",
"type": "module",
"main": "./src/index.ts",
@@ -10,9 +10,21 @@
"scripts": {
"start": "bun --watch --env-file=.env.development src/index.ts",
"build": "bun ../../scripts/build/server.ts --target=all",
"test:tools": "bun run test:cleanup && bun --env-file=.env.development test tests/tools",
"test:integration": "bun run test:cleanup && bun --env-file=.env.development test tests/server.integration.test.ts",
"test:sdk": "echo 'SDK tests disabled: test environment does not provide the extract/verify LLM service'",
"test": "bun run test:all",
"test:all": "bun run ./tests/__helpers__/run-test-group.ts all",
"test:agent": "bun run ./tests/__helpers__/run-test-group.ts agent",
"test:api": "bun run ./tests/__helpers__/run-test-group.ts api",
"test:browser": "bun run ./tests/__helpers__/run-test-group.ts browser",
"test:cdp": "bun run test:browser",
"test:core": "bun run ./tests/__helpers__/run-test-group.ts core",
"test:integration": "bun run ./tests/__helpers__/run-test-group.ts integration",
"test:root": "bun run ./tests/__helpers__/run-test-group.ts root",
"test:sdk": "bun run ./tests/__helpers__/run-test-group.ts sdk",
"test:skills": "bun run ./tests/__helpers__/run-test-group.ts skills",
"test:tools": "bun run ./tests/__helpers__/run-test-group.ts tools",
"test:tools:acl": "bun run test:cleanup && bun --env-file=.env.development test ./tests/tools/acl-scorer.test.ts",
"test:tools:filesystem": "bun run test:cleanup && bun --env-file=.env.development test ./tests/tools/filesystem",
"test:tools:input": "bun run test:cleanup && bun --env-file=.env.development test ./tests/tools/input.test.ts",
"test:cleanup": "./tests/__helpers__/cleanup.sh",
"typecheck": "tsc --noEmit",
"devtools": "bunx @ai-sdk/devtools"

View File

@@ -1,37 +0,0 @@
services:
openclaw-gateway:
image: ${OPENCLAW_IMAGE:-ghcr.io/openclaw/openclaw:latest}
ports:
- "127.0.0.1:${OPENCLAW_GATEWAY_PORT:-18789}:18789"
environment:
- HOME=/home/node
- NODE_ENV=production
- OPENCLAW_GATEWAY_TOKEN=${OPENCLAW_GATEWAY_TOKEN}
- OPENCLAW_GATEWAY_BIND=lan
- TZ=${TZ}
- ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY:-}
- OPENAI_API_KEY=${OPENAI_API_KEY:-}
- GEMINI_API_KEY=${GEMINI_API_KEY:-}
- OPENROUTER_API_KEY=${OPENROUTER_API_KEY:-}
- GROQ_API_KEY=${GROQ_API_KEY:-}
- MISTRAL_API_KEY=${MISTRAL_API_KEY:-}
- MOONSHOT_API_KEY=${MOONSHOT_API_KEY:-}
volumes:
- ${OPENCLAW_CONFIG_DIR}:/home/node/.openclaw
extra_hosts:
- "host.containers.internal:host-gateway"
command:
- node
- dist/index.js
- gateway
- --bind
- lan
- --port
- "18789"
- --allow-unconfigured
healthcheck:
test: ["CMD", "curl", "-sf", "http://127.0.0.1:18789/healthz"]
interval: 30s
timeout: 10s
retries: 3
restart: unless-stopped

View File

@@ -10,6 +10,7 @@ import type { Browser } from '../../browser/browser'
import { logger } from '../../lib/logger'
import { metrics } from '../../lib/metrics'
import { Sentry } from '../../lib/sentry'
import { getMonitoringService } from '../../monitoring/service'
import type { ToolRegistry } from '../../tools/tool-registry'
import type { GlobalAclPolicyService } from '../services/acl/global-acl-policy'
import { resolveAclPolicyForMcpRequest } from '../services/acl/resolve-acl-policy'
@@ -39,16 +40,30 @@ export function createMcpRoutes(deps: McpRouteDeps) {
app.post('/', async (c) => {
const scopeId = c.req.header('X-BrowserOS-Scope-Id') || 'ephemeral'
const monitoringService = getMonitoringService()
const explicitAgentId =
c.req.query('agentId') ??
c.req.header('X-BrowserOS-Agent-Id') ??
undefined
const activeSession =
monitoringService.resolveSessionForMcpRequest(explicitAgentId)
const agentId = activeSession?.agentId
metrics.log('mcp.request', { scopeId })
const aclRules = await resolveAclPolicyForMcpRequest({
policyService: deps.policyService,
})
const monitoringSessionId = activeSession?.monitoringSessionId
const observer =
monitoringSessionId && agentId
? monitoringService.createObserver(monitoringSessionId, agentId)
: undefined
// Per-request server + transport: no shared state, no race conditions,
// no ID collisions. Required by MCP SDK 1.26.0+ security fix (GHSA-345p-7cg4-v4c7).
const mcpServer = createMcpServer({
...deps,
aclRules,
observer,
})
const transport = new StreamableHTTPTransport({
sessionIdGenerator: undefined,
@@ -62,6 +77,9 @@ export function createMcpRoutes(deps: McpRouteDeps) {
Sentry.withScope((scope) => {
scope.setTag('route', 'mcp')
scope.setTag('scopeId', scopeId)
if (agentId) {
scope.setTag('agentId', agentId)
}
Sentry.captureException(error)
})
logger.error('Error handling MCP request', {

View File

@@ -0,0 +1,113 @@
import { Hono } from 'hono'
import { getMonitoringService } from '../../monitoring/service'
import { isValidMonitoringRunId } from '../../monitoring/storage'
export function createMonitoringRoutes() {
return new Hono()
.get('/runs', async (c) => {
const limitParam = c.req.query('limit')
const parsedLimit = limitParam ? Number.parseInt(limitParam, 10) : 50
const limit =
Number.isFinite(parsedLimit) && parsedLimit > 0 ? parsedLimit : 50
const runs = await getMonitoringService().listRuns(limit)
return c.json({ runs })
})
.get('/runs/:id', async (c) => {
const runId = c.req.param('id')
if (!isValidMonitoringRunId(runId)) {
return c.json({ error: 'Invalid monitoring run id' }, 400)
}
const envelope = await getMonitoringService().getRunEnvelope(runId)
if (!envelope) {
return c.json({ error: 'Monitoring run not found' }, 404)
}
return c.json({ run: envelope })
})
.post('/debug/runs', async (c) => {
const body = await c.req.json<{
agentId?: string
sessionKey?: string
originalPrompt?: string
chatHistory?: Array<{ role?: 'user' | 'assistant'; content?: string }>
}>()
if (!body.agentId?.trim()) {
return c.json({ error: 'agentId is required' }, 400)
}
if (!body.sessionKey?.trim()) {
return c.json({ error: 'sessionKey is required' }, 400)
}
if (!body.originalPrompt?.trim()) {
return c.json({ error: 'originalPrompt is required' }, 400)
}
const chatHistory = Array.isArray(body.chatHistory)
? body.chatHistory
.filter(
(turn): turn is { role: 'user' | 'assistant'; content: string } =>
(turn.role === 'user' || turn.role === 'assistant') &&
typeof turn.content === 'string',
)
.map((turn) => ({
role: turn.role,
content: turn.content,
}))
: []
const session = await getMonitoringService().startSession({
agentId: body.agentId.trim(),
sessionKey: body.sessionKey.trim(),
originalPrompt: body.originalPrompt.trim(),
chatHistory,
source: 'debug',
})
return c.json({ session }, 201)
})
.post('/debug/runs/:id/finalize', async (c) => {
const runId = c.req.param('id')
if (!isValidMonitoringRunId(runId)) {
return c.json({ error: 'Invalid monitoring run id' }, 400)
}
const body = await c.req.json<{
agentId?: string
sessionKey?: string
status?: 'completed' | 'failed' | 'aborted' | 'incomplete'
finalAssistantMessage?: string
error?: string
}>()
if (!body.agentId?.trim()) {
return c.json({ error: 'agentId is required' }, 400)
}
if (!body.sessionKey?.trim()) {
return c.json({ error: 'sessionKey is required' }, 400)
}
if (
body.status !== 'completed' &&
body.status !== 'failed' &&
body.status !== 'aborted' &&
body.status !== 'incomplete'
) {
return c.json({ error: 'status is invalid' }, 400)
}
const envelope = await getMonitoringService().finalizeSession({
monitoringSessionId: runId,
agentId: body.agentId.trim(),
sessionKey: body.sessionKey.trim(),
status: body.status,
finalAssistantMessage: body.finalAssistantMessage,
error: body.error,
})
if (!envelope) {
return c.json({ error: 'Monitoring run not found' }, 404)
}
return c.json({ run: envelope })
})
}

View File

@@ -7,38 +7,200 @@
* Thin layer delegating to OpenClawService.
*/
import { OPENCLAW_GATEWAY_PORT } from '@browseros/shared/constants/openclaw'
import { BROWSEROS_ROLE_TEMPLATES } from '@browseros/shared/constants/role-aware-agents'
import type {
BrowserOSAgentRoleId,
BrowserOSCustomRoleInput,
} from '@browseros/shared/types/role-aware-agents'
import { Hono } from 'hono'
import { stream } from 'hono/streaming'
import { logger } from '../../lib/logger'
import { getMonitoringService } from '../../monitoring/service'
import type { MonitoringChatTurn } from '../../monitoring/types'
import {
OpenClawAgentAlreadyExistsError,
OpenClawAgentNotFoundError,
OpenClawInvalidAgentNameError,
OpenClawProtectedAgentError,
OpenClawSessionNotFoundError,
} from '../services/openclaw/errors'
import { getOpenClawService } from '../services/openclaw/openclaw-service'
import { getOpenClawCliProvider } from '../services/openclaw/openclaw-cli-providers/registry'
import type { OpenClawChatContentPart } from '../services/openclaw/openclaw-http-client'
import { isUnsupportedOpenClawProviderError } from '../services/openclaw/openclaw-provider-map'
import {
getOpenClawService,
normalizeBrowserOSChatSessionKey,
} from '../services/openclaw/openclaw-service'
import type { QueuedItemPublic } from '../services/queue'
import { getOutboundQueueService } from '../services/queue'
function isValidBoundaryMode(
value: unknown,
): value is BrowserOSCustomRoleInput['boundaries'][number]['defaultMode'] {
return value === 'allow' || value === 'ask' || value === 'block'
/**
* Inbound attachment shapes the chat route accepts. Images travel as
* data: URLs (the gateway is on 127.0.0.1 so we don't pay public-network
* cost for the base64 overhead). Files arrive with their text already
* extracted on the client — we just inline them as a fenced text part on
* the user message.
*/
type ImageAttachment = {
kind: 'image'
mediaType: string
dataUrl: string
name?: string
}
type FileAttachment = {
kind: 'file'
mediaType: string
name: string
text: string
}
type ChatAttachment = ImageAttachment | FileAttachment
const MAX_ATTACHMENTS = 10
const MAX_IMAGE_BYTES = 5 * 1024 * 1024 // 5 MB after compression
// data: URLs encode bytes as base64 (~4/3 inflation) plus a small media-type
// prefix; cap the encoded string against that, not 2× the byte budget.
const MAX_IMAGE_DATA_URL_LENGTH = Math.ceil(MAX_IMAGE_BYTES * (4 / 3)) + 100
const MAX_FILE_TEXT_BYTES = 1 * 1024 * 1024 // 1 MB extracted text
const ALLOWED_IMAGE_MEDIA_TYPES = new Set([
'image/png',
'image/jpeg',
'image/jpg',
'image/webp',
'image/gif',
])
const ALLOWED_FILE_MEDIA_TYPE_PREFIXES = ['text/', 'application/json']
function validateChatAttachments(input: unknown): {
attachments: ChatAttachment[] | null
error: string | null
} {
if (input === undefined || input === null) {
return { attachments: null, error: null }
}
if (!Array.isArray(input)) {
return { attachments: null, error: 'attachments must be an array' }
}
if (input.length > MAX_ATTACHMENTS) {
return {
attachments: null,
error: `at most ${MAX_ATTACHMENTS} attachments are allowed per message`,
}
}
const result: ChatAttachment[] = []
for (const raw of input) {
if (!raw || typeof raw !== 'object') {
return { attachments: null, error: 'invalid attachment entry' }
}
const entry = raw as Record<string, unknown>
if (entry.kind === 'image') {
const mediaType =
typeof entry.mediaType === 'string' ? entry.mediaType : ''
const dataUrl = typeof entry.dataUrl === 'string' ? entry.dataUrl : ''
if (!ALLOWED_IMAGE_MEDIA_TYPES.has(mediaType)) {
return {
attachments: null,
error: `unsupported image type: ${mediaType || 'unknown'}`,
}
}
if (!dataUrl.startsWith('data:')) {
return {
attachments: null,
error: 'image attachment must include a data: URL',
}
}
if (dataUrl.length > MAX_IMAGE_DATA_URL_LENGTH) {
return {
attachments: null,
error: `image exceeds ${MAX_IMAGE_BYTES} bytes`,
}
}
result.push({
kind: 'image',
mediaType,
dataUrl,
name: typeof entry.name === 'string' ? entry.name : undefined,
})
continue
}
if (entry.kind === 'file') {
const mediaType =
typeof entry.mediaType === 'string' ? entry.mediaType : ''
const name = typeof entry.name === 'string' ? entry.name : ''
const text = typeof entry.text === 'string' ? entry.text : ''
const allowed = ALLOWED_FILE_MEDIA_TYPE_PREFIXES.some((prefix) =>
mediaType.startsWith(prefix),
)
if (!allowed) {
return {
attachments: null,
error: `unsupported file type: ${mediaType || 'unknown'}`,
}
}
if (!name) {
return {
attachments: null,
error: 'file attachment must include a name',
}
}
if (text.length > MAX_FILE_TEXT_BYTES) {
return {
attachments: null,
error: `file "${name}" exceeds ${MAX_FILE_TEXT_BYTES} bytes`,
}
}
result.push({ kind: 'file', mediaType, name, text })
continue
}
return {
attachments: null,
error: 'attachment kind must be "image" or "file"',
}
}
return { attachments: result, error: null }
}
function isValidCustomRoleBoundary(value: unknown): boolean {
if (!value || typeof value !== 'object') return false
const boundary = value as Record<string, unknown>
return (
typeof boundary.key === 'string' &&
typeof boundary.label === 'string' &&
typeof boundary.description === 'string' &&
isValidBoundaryMode(boundary.defaultMode)
function buildMessagePartsFromAttachments(
message: string,
attachments: ChatAttachment[],
): { text: string; parts: OpenClawChatContentPart[] | undefined } {
const images = attachments.filter(
(a): a is ImageAttachment => a.kind === 'image',
)
const files = attachments.filter(
(a): a is FileAttachment => a.kind === 'file',
)
const fileBlocks = files
.map(
(f) => `<attachment name="${f.name}" mediaType="${f.mediaType}">
${f.text}
</attachment>`,
)
.join('\n\n')
const text = fileBlocks ? `${message}\n\n${fileBlocks}`.trim() : message
if (images.length === 0) {
return { text, parts: undefined }
}
const parts: OpenClawChatContentPart[] = [{ type: 'text', text }]
for (const image of images) {
parts.push({ type: 'image_url', image_url: { url: image.dataUrl } })
}
return { text, parts }
}
function getCreateAgentValidationError(body: { name?: string }): string | null {
if (!body.name?.trim()) {
return 'Name is required'
}
return null
}
function parsePositiveIntQuery(
value: string | undefined,
fallback: number,
): number {
if (value === undefined) return fallback
const parsed = Number(value)
if (!Number.isFinite(parsed)) return fallback
return Math.max(1, Math.trunc(parsed))
}
export function createOpenClawRoutes() {
@@ -48,6 +210,29 @@ export function createOpenClawRoutes() {
return c.json(status)
})
.get('/providers/:providerId/auth-status', async (c) => {
const { providerId } = c.req.param()
const provider = getOpenClawCliProvider(providerId)
if (!provider) {
return c.json({ error: `Unknown CLI provider: ${providerId}` }, 404)
}
try {
const status =
await getOpenClawService().getCliProviderAuthStatus(provider)
return c.json(status)
} catch (err) {
const message = err instanceof Error ? err.message : String(err)
logger.warn('CLI provider auth-status failed', {
providerId,
error: message,
})
return c.json(
{ installed: false, loggedIn: false, error: message },
500,
)
}
})
.post('/setup', async (c) => {
const body = await c.req.json<{
providerType?: string
@@ -72,7 +257,7 @@ export function createOpenClawRoutes() {
return c.json(
{
status: 'running',
port: OPENCLAW_GATEWAY_PORT,
port: getOpenClawService().getPort(),
agents: agents.map((a) => ({
agentId: a.agentId,
name: a.name,
@@ -89,7 +274,10 @@ export function createOpenClawRoutes() {
providerType: body.providerType,
providerName: body.providerName,
})
if (message.includes('Podman is not available')) {
if (isUnsupportedOpenClawProviderError(err)) {
return c.json({ error: err.message }, 400)
}
if (message.includes('VM runtime is not available')) {
return c.json({ error: message }, 503)
}
return c.json({ error: message }, 500)
@@ -154,97 +342,23 @@ export function createOpenClawRoutes() {
}
})
.get('/roles', async (c) => {
return c.json({
roles: BROWSEROS_ROLE_TEMPLATES.map((role) => ({
id: role.id,
name: role.name,
shortDescription: role.shortDescription,
longDescription: role.longDescription,
recommendedApps: role.recommendedApps,
boundaries: role.boundaries,
defaultAgentName: role.defaultAgentName,
})),
})
})
.post('/agents', async (c) => {
const body = await c.req.json<{
name: string
roleId?: BrowserOSAgentRoleId
customRole?: BrowserOSCustomRoleInput
providerType?: string
providerName?: string
baseUrl?: string
apiKey?: string
modelId?: string
}>()
const name = body.name?.trim()
if (!name) {
return c.json({ error: 'Name is required' }, 400)
}
if (body.roleId && body.customRole) {
return c.json(
{ error: 'Provide either roleId or customRole, not both' },
400,
)
}
if (
body.customRole &&
(!body.customRole.name?.trim() ||
!body.customRole.shortDescription?.trim() ||
!body.customRole.longDescription?.trim())
) {
return c.json(
{
error:
'Custom roles require name, shortDescription, and longDescription',
},
400,
)
}
if (
body.customRole &&
(!Array.isArray(body.customRole.recommendedApps) ||
!Array.isArray(body.customRole.boundaries))
) {
return c.json(
{
error: 'Custom roles require recommendedApps and boundaries arrays',
},
400,
)
}
if (
body.customRole &&
!body.customRole.recommendedApps.every((app) => typeof app === 'string')
) {
return c.json(
{
error: 'Custom role recommendedApps must be an array of strings',
},
400,
)
}
if (
body.customRole &&
!body.customRole.boundaries.every(isValidCustomRoleBoundary)
) {
return c.json(
{
error:
'Custom role boundaries must include key, label, description, and a valid defaultMode',
},
400,
)
const validationError = getCreateAgentValidationError(body)
if (validationError) {
return c.json({ error: validationError }, 400)
}
try {
const agent = await getOpenClawService().createAgent({
name,
roleId: body.roleId,
customRole: body.customRole,
name: body.name.trim(),
providerType: body.providerType,
providerName: body.providerName,
baseUrl: body.baseUrl,
@@ -259,6 +373,9 @@ export function createOpenClawRoutes() {
if (err instanceof OpenClawInvalidAgentNameError) {
return c.json({ error: err.message }, 400)
}
if (isUnsupportedOpenClawProviderError(err)) {
return c.json({ error: err.message }, 400)
}
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
@@ -282,24 +399,189 @@ export function createOpenClawRoutes() {
}
})
.get('/agents/:id/sessions', async (c) => {
const { id } = c.req.param()
const limit = parsePositiveIntQuery(c.req.query('limit'), 20)
try {
const sessions = await getOpenClawService().listSessions(id)
return c.json({
agentId: id,
sessions: sessions.slice(0, Math.min(limit, 100)),
})
} catch (err) {
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
})
.get('/agents/:id/session', async (c) => {
const { id } = c.req.param()
try {
const session = await getOpenClawService().resolveAgentSession(id)
return c.json(session)
} catch (err) {
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
})
.get('/agents/:id/history', async (c) => {
const { id } = c.req.param()
const limit = parsePositiveIntQuery(c.req.query('limit'), 50)
try {
const page = await getOpenClawService().getAgentHistoryPage(id, {
sessionKey: c.req.query('sessionKey'),
cursor: c.req.query('cursor'),
limit,
})
return c.json(page)
} catch (err) {
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
})
.get('/dashboard', (c) => {
try {
const dashboard = getOpenClawService().getDashboard()
return c.json(dashboard)
} catch (err) {
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
})
.get('/dashboard/stream', (c) => {
c.header('Content-Type', 'text/event-stream')
c.header('Cache-Control', 'no-cache')
c.header('Connection', 'keep-alive')
return stream(c, async (s) => {
const encoder = new TextEncoder()
// Send initial snapshot
try {
const dashboard = getOpenClawService().getDashboard()
await s.write(
encoder.encode(
`event: snapshot\ndata: ${JSON.stringify(dashboard)}\n\n`,
),
)
} catch {}
// Subscribe to live status changes
const unsubscribe = getOpenClawService().onAgentStatusChange(
(agentId, entry) => {
const event = {
agentId,
status: entry.status,
currentTool: entry.currentTool,
error: entry.error,
timestamp: entry.lastEventAt,
}
s.write(
encoder.encode(
`event: status\ndata: ${JSON.stringify(event)}\n\n`,
),
).catch(() => {})
},
)
// Heartbeat every 15s to keep connection alive
const heartbeat = setInterval(() => {
s.write(
encoder.encode(
`event: heartbeat\ndata: ${JSON.stringify({ ts: Date.now() })}\n\n`,
),
).catch(() => {})
}, 15_000)
// Wait until client disconnects
try {
await new Promise<void>((resolve) => {
s.onAbort(() => resolve())
})
} finally {
unsubscribe()
clearInterval(heartbeat)
}
})
})
.post('/agents/:id/chat', async (c) => {
const { id } = c.req.param()
const body = await c.req.json<{
message: string
sessionKey?: string
history?: MonitoringChatTurn[]
attachments?: unknown
}>()
if (!body.message?.trim()) {
const trimmedMessage = body.message?.trim() ?? ''
const attachmentValidation = validateChatAttachments(body.attachments)
if (attachmentValidation.error) {
return c.json({ error: attachmentValidation.error }, 400)
}
const attachments = attachmentValidation.attachments ?? []
// Either a non-empty text body or at least one attachment is required.
if (!trimmedMessage && attachments.length === 0) {
return c.json({ error: 'Message is required' }, 400)
}
const sessionKey = body.sessionKey ?? crypto.randomUUID()
const sessionKey = normalizeBrowserOSChatSessionKey(
id,
body.sessionKey ?? crypto.randomUUID(),
)
const history = Array.isArray(body.history)
? body.history.filter((entry): entry is MonitoringChatTurn =>
Boolean(
entry &&
(entry.role === 'user' || entry.role === 'assistant') &&
typeof entry.content === 'string',
),
)
: []
// Replace the immediate 409 with a bounded wait so back-to-back user
// sends or a cron / hook turn that's still finishing don't reject the
// user-chat outright. The client-side outbound queue (Feature 2) keeps
// the per-agent send rate at 1, so this only kicks in for cross-source
// contention.
try {
await getMonitoringService().waitForSessionFree(id, {
timeoutMs: 30_000,
})
} catch (err) {
return c.json(
{
error:
err instanceof Error
? err.message
: 'Agent is busy. Try again shortly.',
},
503,
)
}
const { text: composedMessage, parts: messageParts } =
buildMessagePartsFromAttachments(trimmedMessage, attachments)
const monitoringContext = await getMonitoringService().startSession({
agentId: id,
sessionKey,
originalPrompt: composedMessage,
chatHistory: history,
})
try {
const eventStream = await getOpenClawService().chatStream(
id,
sessionKey,
body.message,
composedMessage,
history,
{ messageParts },
)
c.header('Content-Type', 'text/event-stream')
@@ -309,20 +591,227 @@ export function createOpenClawRoutes() {
return stream(c, async (s) => {
const reader = eventStream.getReader()
const encoder = new TextEncoder()
let finalAssistantMessage: string | undefined
let status: 'completed' | 'failed' | 'aborted' | 'incomplete' =
'incomplete'
let finalError: string | undefined
try {
while (true) {
const { done, value } = await reader.read()
if (done) break
if (
value.type === 'done' &&
typeof value.data.text === 'string' &&
value.data.text.trim()
) {
finalAssistantMessage = value.data.text
status = 'completed'
}
if (value.type === 'error') {
finalError =
(typeof value.data.message === 'string'
? value.data.message
: typeof value.data.error === 'string'
? value.data.error
: undefined) ?? 'Unknown chat stream error'
status = 'failed'
}
await s.write(
encoder.encode(`data: ${JSON.stringify(value)}\n\n`),
)
}
await s.write(encoder.encode('data: [DONE]\n\n'))
} catch (error) {
if (c.req.raw.signal.aborted) {
status = 'aborted'
} else {
status = 'failed'
finalError =
error instanceof Error ? error.message : String(error)
}
throw error
} finally {
await reader.cancel()
await getMonitoringService().finalizeSession({
monitoringSessionId: monitoringContext.monitoringSessionId,
agentId: id,
sessionKey,
status,
finalAssistantMessage,
error: finalError,
})
}
})
} catch (err) {
await getMonitoringService().finalizeSession({
monitoringSessionId: monitoringContext.monitoringSessionId,
agentId: id,
sessionKey,
status: c.req.raw.signal.aborted ? 'aborted' : 'failed',
error: err instanceof Error ? err.message : String(err),
})
if (isUnsupportedOpenClawProviderError(err)) {
return c.json({ error: err.message }, 400)
}
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
})
.post('/agents/:id/queue', async (c) => {
const { id } = c.req.param()
const body = await c.req.json<{
message: string
sessionKey?: string
history?: MonitoringChatTurn[]
attachments?: unknown
// Optional client-provided id — when set, the queue uses it as
// the canonical item id so the browser's optimistic row and the
// SSE snapshot reconcile on the same key.
id?: string
}>()
const trimmedMessage = body.message?.trim() ?? ''
const attachmentValidation = validateChatAttachments(body.attachments)
if (attachmentValidation.error) {
return c.json({ error: attachmentValidation.error }, 400)
}
const attachments = attachmentValidation.attachments ?? []
if (!trimmedMessage && attachments.length === 0) {
return c.json({ error: 'Message is required' }, 400)
}
const sessionKey = body.sessionKey
? normalizeBrowserOSChatSessionKey(id, body.sessionKey)
: undefined
const history = Array.isArray(body.history)
? body.history.filter((entry): entry is MonitoringChatTurn =>
Boolean(
entry &&
(entry.role === 'user' || entry.role === 'assistant') &&
typeof entry.content === 'string',
),
)
: []
const { text: composedMessage, parts: messageParts } =
buildMessagePartsFromAttachments(trimmedMessage, attachments)
const item = getOutboundQueueService().enqueue({
agentId: id,
id: typeof body.id === 'string' && body.id ? body.id : undefined,
message: composedMessage,
messageParts,
sessionKey,
history,
attachmentsPreview: attachments.map((a) => ({
kind: a.kind,
mediaType: a.mediaType,
name: 'name' in a ? a.name : undefined,
})),
})
return c.json({ id: item.id }, 202)
})
.delete('/agents/:id/queue/:itemId', (c) => {
const { id, itemId } = c.req.param()
const result = getOutboundQueueService().cancel(id, itemId)
if (!result.ok) {
const code = result.reason === 'dispatching' ? 409 : 404
const message =
result.reason === 'dispatching'
? 'Item is already dispatching'
: 'Item not found'
return c.json({ error: message }, code)
}
return c.json({ ok: true })
})
.post('/agents/:id/queue/:itemId/retry', (c) => {
const { id, itemId } = c.req.param()
const result = getOutboundQueueService().retry(id, itemId)
if (!result.ok) {
return c.json({ error: 'Item not found or not failed' }, 404)
}
return c.json({ ok: true })
})
.get('/agents/:id/queue/stream', (c) => {
const { id } = c.req.param()
c.header('Content-Type', 'text/event-stream')
c.header('Cache-Control', 'no-cache')
return stream(c, async (s) => {
const encoder = new TextEncoder()
const sendSnapshot = (items: QueuedItemPublic[]) => {
void s.write(encoder.encode(`data: ${JSON.stringify({ items })}\n\n`))
}
const unsubscribe = getOutboundQueueService().subscribe(
id,
sendSnapshot,
)
const heartbeat = setInterval(() => {
void s.write(encoder.encode(': keep-alive\n\n'))
}, 15_000)
try {
await new Promise<void>((resolve) => {
s.onAbort(() => resolve())
})
} finally {
clearInterval(heartbeat)
unsubscribe()
}
})
})
.get('/session/:key/history', async (c) => {
const key = c.req.param('key')
const limitRaw = c.req.query('limit')
const cursor = c.req.query('cursor')
const limitParsed =
limitRaw !== undefined ? Number.parseInt(limitRaw, 10) : Number.NaN
const limit = Number.isFinite(limitParsed) ? limitParsed : undefined
const wantsStream = (c.req.header('accept') ?? '').includes(
'text/event-stream',
)
try {
if (!wantsStream) {
const history = await getOpenClawService().getSessionHistory(key, {
limit,
cursor,
})
return c.json(history)
}
const eventStream = await getOpenClawService().streamSessionHistory(
key,
{ limit, cursor, signal: c.req.raw.signal },
)
c.header('Content-Type', 'text/event-stream')
c.header('Cache-Control', 'no-cache')
c.header('X-Session-Key', key)
return stream(c, async (s) => {
const reader = eventStream.getReader()
const encoder = new TextEncoder()
try {
while (true) {
const { done, value } = await reader.read()
if (done) break
await s.write(
encoder.encode(
`event: ${value.type}\ndata: ${JSON.stringify(value.data)}\n\n`,
),
)
}
} finally {
await reader.cancel()
}
})
} catch (err) {
if (err instanceof OpenClawSessionNotFoundError) {
return c.json({ error: err.message }, 404)
}
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}
@@ -352,12 +841,17 @@ export function createOpenClawRoutes() {
}
try {
await getOpenClawService().updateProviderKeys(body)
const result = await getOpenClawService().updateProviderKeys(body)
return c.json({
status: 'restarting',
message: 'Provider updated, restarting gateway',
status: result.restarted ? 'restarting' : 'updated',
message: result.restarted
? 'Provider updated, restarting gateway'
: 'Provider updated without a restart',
})
} catch (err) {
if (isUnsupportedOpenClawProviderError(err)) {
return c.json({ error: err.message }, 400)
}
const message = err instanceof Error ? err.message : String(err)
return c.json({ error: message }, 500)
}

View File

@@ -16,7 +16,9 @@ export const TERMINAL_WS_PATH = '/terminal/ws'
interface TerminalRouteDeps {
containerName: string
podmanPath: string
limaHome: string
limactlPath: string
vmName: string
}
function safeSend(ws: { send(data: string): void }, data: string): void {
@@ -45,7 +47,9 @@ function createSocketEvents(deps: TerminalRouteDeps) {
try {
session = createTerminalSession({
containerName: deps.containerName,
podmanPath: deps.podmanPath,
limaHome: deps.limaHome,
limactlPath: deps.limactlPath,
vmName: deps.vmName,
workingDir: TERMINAL_HOME_DIR,
onOutput(data) {
sendOutput(ws, data)

View File

@@ -22,6 +22,7 @@ import { initializeOAuth } from '../lib/clients/oauth'
import { getDb } from '../lib/db'
import { logger } from '../lib/logger'
import { Sentry } from '../lib/sentry'
import { getLimaHomeDir, resolveBundledLimactl, VM_NAME } from '../lib/vm'
import { createAclRoutes } from './routes/acl'
import { createChatRoutes } from './routes/chat'
import { createCreditsRoutes } from './routes/credits'
@@ -29,6 +30,7 @@ import { createHealthRoute } from './routes/health'
import { createKlavisRoutes } from './routes/klavis'
import { createMcpRoutes } from './routes/mcp'
import { createMemoryRoutes } from './routes/memory'
import { createMonitoringRoutes } from './routes/monitoring'
import { createOAuthRoutes } from './routes/oauth'
import { createOpenClawRoutes } from './routes/openclaw'
import { createProviderRoutes } from './routes/provider'
@@ -44,7 +46,6 @@ import {
connectKlavisInBackground,
type KlavisProxyRef,
} from './services/klavis/strata-proxy'
import { getPodmanRuntime } from './services/openclaw/podman-runtime'
import type { Env, HttpServerConfig } from './types'
import { defaultCorsConfig } from './utils/cors'
import { requireTrustedAppOrigin } from './utils/request-auth'
@@ -113,7 +114,9 @@ export async function createHttpServer(config: HttpServerConfig) {
'/',
createTerminalRoutes({
containerName: OPENCLAW_GATEWAY_CONTAINER_NAME,
podmanPath: getPodmanRuntime().getPodmanPath(),
limaHome: getLimaHomeDir(),
limactlPath: resolveBundledLimactl(resourcesDir),
vmName: VM_NAME,
}),
)
@@ -121,6 +124,10 @@ export async function createHttpServer(config: HttpServerConfig) {
.use('/*', requireTrustedAppOrigin())
.route('/', createAclRoutes({ policyService: aclPolicyService }))
const monitoringRoutes = new Hono<Env>()
.use('/*', requireTrustedAppOrigin())
.route('/', createMonitoringRoutes())
const app = new Hono<Env>()
.use('/*', cors(defaultCorsConfig))
.route('/health', createHealthRoute({ browser }))
@@ -143,6 +150,7 @@ export async function createHttpServer(config: HttpServerConfig) {
.route('/soul', createSoulRoutes())
.route('/memory', createMemoryRoutes())
.route('/skills', createSkillsRoutes())
.route('/monitoring', monitoringRoutes)
.route('/acl-rules', aclRoutes)
.route('/test-provider', createProviderRoutes({ browserosId }))
.route('/refine-prompt', createRefinePromptRoutes({ browserosId }))

View File

@@ -20,6 +20,10 @@ import { KlavisClient } from '../../../lib/clients/klavis/klavis-client'
import { OAUTH_MCP_SERVERS } from '../../../lib/clients/klavis/oauth-mcp-servers'
import { logger } from '../../../lib/logger'
import { metrics } from '../../../lib/metrics'
import {
buildMonitoringToolOutput,
type ToolExecutionObserver,
} from '../../../monitoring/observer'
import { klavisStrataCache } from './strata-cache'
function withTimeout<T>(promise: Promise<T>, label: string): Promise<T> {
@@ -237,6 +241,7 @@ export function buildKlavisToolSet(handle: KlavisProxyHandle): ToolSet {
export function registerKlavisTools(
mcpServer: McpServer,
handle: KlavisProxyHandle,
observer?: ToolExecutionObserver,
): void {
mcpServer.registerTool(
'connector_mcp_servers',
@@ -247,9 +252,18 @@ export function registerKlavisTools(
},
async (args: Record<string, unknown>) => {
const startTime = performance.now()
const toolCallId = crypto.randomUUID()
const server_name = args.server_name as string
try {
await observer?.onToolStart({
toolCallId,
toolName: 'connector_mcp_servers',
toolDescription:
'Check whether an external connector is connected and ready for use.',
source: 'klavis-tool',
args,
})
const klavisClient = new KlavisClient()
const integrations = await klavisClient.getUserIntegrations(
handle.browserosId,
@@ -266,6 +280,14 @@ export function registerKlavisTools(
success: true,
})
await observer?.onToolEnd({
toolCallId,
output: {
connected: true,
server_name,
},
})
return {
content: [
{
@@ -294,6 +316,15 @@ export function registerKlavisTools(
success: true,
})
await observer?.onToolEnd({
toolCallId,
output: {
connected: false,
server_name,
authUrl,
},
})
return {
content: [
{
@@ -320,6 +351,11 @@ export function registerKlavisTools(
error_message: errorText,
})
await observer?.onToolEnd({
toolCallId,
error: errorText,
})
return {
content: [{ type: 'text' as const, text: errorText }],
isError: true,
@@ -339,7 +375,15 @@ export function registerKlavisTools(
},
async (args: Record<string, unknown>) => {
const startTime = performance.now()
const toolCallId = crypto.randomUUID()
try {
await observer?.onToolStart({
toolCallId,
toolName: tool.name,
toolDescription: tool.description ?? undefined,
source: 'klavis-tool',
args,
})
const result = await handle.callTool(tool.name, args)
metrics.log('tool_executed', {
@@ -349,6 +393,12 @@ export function registerKlavisTools(
success: !result.isError,
})
await observer?.onToolEnd({
toolCallId,
output: buildMonitoringToolOutput(result),
error: result.isError ? 'Tool returned isError=true' : undefined,
})
return result
} catch (error) {
const errorText =
@@ -362,6 +412,11 @@ export function registerKlavisTools(
error_message: errorText,
})
await observer?.onToolEnd({
toolCallId,
error: errorText,
})
return {
content: [{ type: 'text' as const, text: errorText }],
isError: true,

View File

@@ -8,6 +8,7 @@ import type { AclRule } from '@browseros/shared/types/acl'
import { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
import { SetLevelRequestSchema } from '@modelcontextprotocol/sdk/types.js'
import type { Browser } from '../../../browser/browser'
import type { ToolExecutionObserver } from '../../../monitoring/observer'
import type { ToolRegistry } from '../../../tools/tool-registry'
import {
type KlavisProxyRef,
@@ -24,6 +25,7 @@ export interface McpServiceDeps {
resourcesDir: string
aclRules?: AclRule[]
klavisRef?: KlavisProxyRef
observer?: ToolExecutionObserver
}
export function createMcpServer(deps: McpServiceDeps): McpServer {
@@ -48,11 +50,12 @@ export function createMcpServer(deps: McpServiceDeps): McpServer {
resourcesDir: deps.resourcesDir,
},
aclRules: deps.aclRules,
observer: deps.observer,
})
// Register Klavis proxy tools (if connected via background init)
if (deps.klavisRef?.handle) {
registerKlavisTools(server, deps.klavisRef.handle)
registerKlavisTools(server, deps.klavisRef.handle, deps.observer)
}
return server

View File

@@ -1,13 +1,17 @@
import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
import { logger } from '../../../lib/logger'
import { metrics } from '../../../lib/metrics'
import {
buildMonitoringToolOutput,
type ToolExecutionObserver,
} from '../../../monitoring/observer'
import { executeTool, type ToolContext } from '../../../tools/framework'
import type { ToolRegistry } from '../../../tools/tool-registry'
export function registerTools(
mcpServer: McpServer,
registry: ToolRegistry,
ctx: ToolContext,
ctx: ToolContext & { observer?: ToolExecutionObserver },
): void {
for (const tool of registry.all()) {
const handler = async (
@@ -15,9 +19,17 @@ export function registerTools(
extra: { signal: AbortSignal },
) => {
const startTime = performance.now()
const toolCallId = crypto.randomUUID()
try {
logger.info(`${tool.name} request: ${JSON.stringify(args, null, ' ')}`)
await ctx.observer?.onToolStart({
toolCallId,
toolName: tool.name,
toolDescription: tool.description,
source: 'browser-tool',
args,
})
const result = await executeTool(tool, args, ctx, extra.signal)
@@ -28,6 +40,17 @@ export function registerTools(
source: 'mcp',
})
await ctx.observer?.onToolEnd({
toolCallId,
output: buildMonitoringToolOutput({
content: result.content,
structuredContent: result.structuredContent,
metadata: result.metadata,
isError: result.isError,
}),
error: result.isError ? 'Tool returned isError=true' : undefined,
})
return {
content: result.content,
isError: result.isError,
@@ -44,6 +67,11 @@ export function registerTools(
source: 'mcp',
})
await ctx.observer?.onToolEnd({
toolCallId,
error: errorText,
})
return {
content: [{ type: 'text' as const, text: errorText }],
isError: true,

View File

@@ -0,0 +1,267 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* In-memory state machine tracking the live status of every OpenClaw agent
* session. Acts as the single source of truth for "is agent X running?"
*
* Two data sources feed it:
* 1. JSONL files (seed) — on init, reads the latest events for each agent
* to infer whether a session is running or idle. This handles the case
* where an agent was already mid-task when BrowserOS started.
* 2. Gateway WS events (live) — the OpenClawObserver pipes chat broadcast
* events into this state machine for real-time transitions.
*
* Consumers (SSE streams, dashboard endpoint) read from this class and get
* correct state from the first call — no "unknown" period while waiting for
* the first WS event.
*/
import { logger } from '../../../lib/logger'
import type { ClawEvent, OpenClawJsonlReader } from './openclaw-jsonl-reader'
// ---------------------------------------------------------------------------
// Types
// ---------------------------------------------------------------------------
export type AgentLiveStatus = 'working' | 'idle' | 'error' | 'unknown'
export interface AgentSessionState {
status: AgentLiveStatus
sessionKey: string | null
lastEventAt: number
currentTool: string | null
error: string | null
}
export type SessionStateListener = (
agentId: string,
state: AgentSessionState,
) => void
// ---------------------------------------------------------------------------
// State machine
// ---------------------------------------------------------------------------
export class ClawSession {
private readonly states = new Map<string, AgentSessionState>()
private readonly listeners = new Set<SessionStateListener>()
private seeded = false
/**
* Seed the state machine from JSONL files. Call this once when the
* gateway becomes ready. For each agent, reads the latest session's
* events and infers whether the agent is currently working or idle.
*
* A session is considered "working" if:
* - The last message-type event is a user.message (agent hasn't replied yet)
* - The last event is an agent.tool_use without a matching agent.tool_result
*
* Otherwise it's "idle".
*/
seedFromJsonl(reader: OpenClawJsonlReader): void {
const agents = reader.listAgents()
for (const agentId of agents) {
const sessions = reader.listSessions(agentId)
if (sessions.length === 0) continue
const latestSession = sessions[0]
const events = reader.listBySession(agentId, latestSession.key)
const state = inferStateFromEvents(events, latestSession.key)
this.states.set(agentId, state)
if (state.status === 'working') {
logger.info('ClawSession seed: agent is working', {
agentId,
currentTool: state.currentTool,
})
}
}
this.seeded = true
logger.info('ClawSession seeded from JSONL', {
agentCount: agents.length,
working: [...this.states.values()].filter((s) => s.status === 'working')
.length,
})
}
/** Whether seedFromJsonl() has been called. */
isSeeded(): boolean {
return this.seeded
}
/** Get the current state of an agent. */
getState(agentId: string): AgentSessionState {
return (
this.states.get(agentId) ?? {
status: 'unknown',
sessionKey: null,
lastEventAt: 0,
currentTool: null,
error: null,
}
)
}
/** Get all tracked agent states. */
getAllStates(): Map<string, AgentSessionState> {
return this.states
}
/**
* Transition an agent's state. Called by the OpenClawObserver when
* a chat WS event arrives.
*/
transition(
agentId: string,
status: AgentLiveStatus,
update: {
sessionKey?: string | null
currentTool?: string | null
error?: string | null
} = {},
): void {
const prev = this.states.get(agentId)
const entry: AgentSessionState = {
status,
sessionKey: update.sessionKey ?? prev?.sessionKey ?? null,
lastEventAt: Date.now(),
currentTool:
status === 'working'
? (update.currentTool ?? prev?.currentTool ?? null)
: null,
error: status === 'error' ? (update.error ?? null) : null,
}
this.states.set(agentId, entry)
for (const listener of this.listeners) {
try {
listener(agentId, entry)
} catch {}
}
}
/** Subscribe to state changes. Returns unsubscribe function. */
onStateChange(listener: SessionStateListener): () => void {
this.listeners.add(listener)
return () => this.listeners.delete(listener)
}
}
// ---------------------------------------------------------------------------
// JSONL state inference
// ---------------------------------------------------------------------------
/**
* Infer the current session state from JSONL events.
*
* The key insight: if the last meaningful event in the JSONL is a
* user.message with no subsequent agent.message, the agent is still
* processing (working). Similarly, an agent.tool_use without a matching
* agent.tool_result means the agent is mid-tool-call.
*
* We also check event recency — if the last event was more than 5 minutes
* ago, we assume the session is idle regardless (handles cases where the
* agent crashed without writing a final event).
*/
function inferStateFromEvents(
events: ClawEvent[],
sessionKey: string,
): AgentSessionState {
if (events.length === 0) {
return {
status: 'idle',
sessionKey,
lastEventAt: 0,
currentTool: null,
error: null,
}
}
const lastEvent = events[events.length - 1]!
const lastEventAt = lastEvent.createdAt
// If the last event is older than 5 minutes, assume idle — the agent
// likely finished or crashed without writing a final event.
const STALE_THRESHOLD_MS = 5 * 60 * 1000
if (Date.now() - lastEventAt > STALE_THRESHOLD_MS) {
return {
status: 'idle',
sessionKey,
lastEventAt,
currentTool: null,
error: null,
}
}
// Walk backward to find the last meaningful event
let lastUserMessageIdx = -1
let lastAssistantMessageIdx = -1
let lastToolUseIdx = -1
let lastToolResultIdx = -1
for (let i = events.length - 1; i >= 0; i--) {
const e = events[i]!
if (e.type === 'user.message' && lastUserMessageIdx === -1) {
lastUserMessageIdx = i
}
if (e.type === 'agent.message' && lastAssistantMessageIdx === -1) {
lastAssistantMessageIdx = i
}
if (e.type === 'agent.tool_use' && lastToolUseIdx === -1) {
lastToolUseIdx = i
}
if (e.type === 'agent.tool_result' && lastToolResultIdx === -1) {
lastToolResultIdx = i
}
// Stop scanning once we've found all event types
if (
lastUserMessageIdx !== -1 &&
lastAssistantMessageIdx !== -1 &&
lastToolUseIdx !== -1 &&
lastToolResultIdx !== -1
) {
break
}
}
// Agent is working if the last user message came AFTER the last
// assistant message — the agent hasn't replied yet
if (
lastUserMessageIdx !== -1 &&
lastUserMessageIdx > lastAssistantMessageIdx
) {
return {
status: 'working',
sessionKey,
lastEventAt,
currentTool: null,
error: null,
}
}
// Agent is working if there's a tool_use without a subsequent tool_result
if (lastToolUseIdx !== -1 && lastToolUseIdx > lastToolResultIdx) {
const toolEvent = events[lastToolUseIdx]!
return {
status: 'working',
sessionKey,
lastEventAt,
currentTool: toolEvent.toolName ?? null,
error: null,
}
}
return {
status: 'idle',
sessionKey,
lastEventAt,
currentTool: null,
error: null,
}
}

View File

@@ -0,0 +1,229 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { cpSync, existsSync, mkdirSync } from 'node:fs'
import { dirname, join } from 'node:path'
import { getBrowserosDir } from '../../../lib/browseros-dir'
import { ContainerCli, ImageLoader } from '../../../lib/container'
import { logger } from '../../../lib/logger'
import {
detectArch,
getLimaHomeDir,
resolveBundledLimactl,
resolveBundledLimaTemplate,
VM_NAME,
VmRuntime,
} from '../../../lib/vm'
import {
ensureVmCacheAvailable,
ensureVmCacheSynced,
type VmCacheSyncOptions,
} from '../../../lib/vm/cache-sync'
import { readCachedManifest } from '../../../lib/vm/manifest'
import { VM_TELEMETRY_EVENTS } from '../../../lib/vm/telemetry'
import { ContainerRuntime } from './container-runtime'
const UNSUPPORTED_PLATFORM_MESSAGE =
'browseros-vm currently supports macOS only; see the Linux/Windows tracking issue'
export interface ContainerRuntimeFactoryInput {
resourcesDir?: string
projectDir: string
browserosRoot?: string
platform?: NodeJS.Platform
vmCache?: VmCacheRuntimeConfig
}
export interface VmCacheRuntimeConfig
extends Pick<VmCacheSyncOptions, 'manifestUrl'> {
ensureAvailable?: () => Promise<void>
ensureSynced?: () => Promise<unknown>
}
export function buildContainerRuntime(
input: ContainerRuntimeFactoryInput,
): ContainerRuntime {
const platform = input.platform ?? process.platform
if (platform !== 'darwin') {
if (process.env.NODE_ENV === 'test') {
return new UnsupportedPlatformTestRuntime(input.projectDir)
}
throw unsupportedPlatformError()
}
const browserosRoot = input.browserosRoot ?? getBrowserosDir()
if (input.resourcesDir) {
migrateLegacyOpenClawDirSync(browserosRoot)
}
const limactlPath = input.resourcesDir
? resolveBundledLimactl(input.resourcesDir)
: 'limactl'
const limaHome = getLimaHomeDir(browserosRoot)
const vm = new VmRuntime({
limactlPath,
limaHome,
templatePath: input.resourcesDir
? resolveBundledLimaTemplate(input.resourcesDir)
: undefined,
browserosRoot,
ensureCacheAvailable:
input.vmCache?.ensureAvailable ??
(() =>
ensureVmCacheAvailable({
browserosRoot,
manifestUrl: input.vmCache?.manifestUrl,
})),
})
const shell = new ContainerCli({ limactlPath, limaHome, vmName: VM_NAME })
const loader = new DeferredImageLoader(shell, browserosRoot, input.vmCache)
return new ContainerRuntime({
vm,
shell,
loader,
projectDir: input.projectDir,
})
}
export async function migrateLegacyOpenClawDir(
browserosRoot = getBrowserosDir(),
): Promise<void> {
migrateLegacyOpenClawDirSync(browserosRoot)
}
function migrateLegacyOpenClawDirSync(browserosRoot = getBrowserosDir()): void {
const legacyDir = join(browserosRoot, 'openclaw')
const nextDir = join(browserosRoot, 'vm', 'openclaw')
if (!existsSync(legacyDir)) return
if (existsSync(nextDir)) {
logger.warn('OpenClaw legacy and VM state directories both exist', {
legacyDir,
nextDir,
})
return
}
mkdirSync(dirname(nextDir), { recursive: true })
cpSync(legacyDir, nextDir, { recursive: true })
logger.info(VM_TELEMETRY_EVENTS.migrationOpenClawMoved, {
from: legacyDir,
to: nextDir,
})
}
class DeferredImageLoader {
constructor(
private readonly shell: ContainerCli,
private readonly browserosRoot: string,
private readonly vmCache?: VmCacheRuntimeConfig,
) {}
async ensureImageLoaded(ref: string, onLog?: (msg: string) => void) {
await this.ensureCacheSynced()
const manifest = await readCachedManifest(this.browserosRoot)
const loader = new ImageLoader(
this.shell,
manifest,
detectArch(),
this.browserosRoot,
)
await loader.ensureImageLoaded(ref, onLog)
}
private async ensureCacheSynced(): Promise<void> {
if (this.vmCache?.ensureSynced) {
await this.vmCache.ensureSynced()
return
}
await ensureVmCacheSynced({
browserosRoot: this.browserosRoot,
manifestUrl: this.vmCache?.manifestUrl,
})
}
}
class UnsupportedPlatformTestRuntime extends ContainerRuntime {
constructor(projectDir: string) {
super({
vm: {} as VmRuntime,
shell: {} as ContainerCli,
loader: { ensureImageLoaded: rejectUnsupportedPlatform },
projectDir,
})
}
override async ensureReady(): Promise<void> {
throw unsupportedPlatformError()
}
override async isPodmanAvailable(): Promise<boolean> {
return false
}
override async getMachineStatus(): Promise<{
initialized: boolean
running: boolean
}> {
return { initialized: false, running: false }
}
override async pullImage(): Promise<void> {
throw unsupportedPlatformError()
}
override async startGateway(): Promise<void> {
throw unsupportedPlatformError()
}
override async stopGateway(): Promise<void> {}
override async restartGateway(): Promise<void> {
throw unsupportedPlatformError()
}
override async getGatewayLogs(): Promise<string[]> {
return []
}
override async isHealthy(): Promise<boolean> {
return false
}
override async isReady(): Promise<boolean> {
return false
}
override async waitForReady(): Promise<boolean> {
return false
}
override async stopVm(): Promise<void> {}
override async execInContainer(): Promise<number> {
throw unsupportedPlatformError()
}
override async runInContainer(): Promise<never> {
throw unsupportedPlatformError()
}
override async runGatewaySetupCommand(): Promise<number> {
throw unsupportedPlatformError()
}
override tailGatewayLogs(): () => void {
return () => {}
}
}
async function rejectUnsupportedPlatform(): Promise<never> {
throw unsupportedPlatformError()
}
function unsupportedPlatformError(): Error {
return new Error(UNSUPPORTED_PLATFORM_MESSAGE)
}

View File

@@ -2,191 +2,316 @@
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Compose-level abstraction over PodmanRuntime.
* Manages a single compose project for the OpenClaw gateway container.
*/
import { copyFile, writeFile } from 'node:fs/promises'
import { join } from 'node:path'
import {
OPENCLAW_COMPOSE_PROJECT_NAME,
OPENCLAW_GATEWAY_CONTAINER_NAME,
OPENCLAW_GATEWAY_CONTAINER_PORT,
} from '@browseros/shared/constants/openclaw'
import type {
ContainerCli,
ContainerCommandResult,
ContainerSpec,
LogFn,
} from '../../../lib/container'
import { logger } from '../../../lib/logger'
import type { LogFn, PodmanRuntime } from './podman-runtime'
import {
GUEST_VM_STATE,
hostPathToGuest,
type VmRuntime,
} from '../../../lib/vm'
const COMPOSE_FILE_NAME = 'docker-compose.yml'
const ENV_FILE_NAME = '.env'
const GATEWAY_CONTAINER_HOME = '/home/node'
const GATEWAY_STATE_DIR = `${GATEWAY_CONTAINER_HOME}/.openclaw`
const GUEST_OPENCLAW_HOME = `${GUEST_VM_STATE}/openclaw`
const GATEWAY_NPM_PREFIX = `${GATEWAY_CONTAINER_HOME}/.npm-global`
// Prepend user-installed bin so tools like `claude` / `gemini` CLI that
// are installed via npm into the mounted home are discoverable by
// OpenClaw's child-process spawns (no login shell is involved).
const GATEWAY_PATH = [
`${GATEWAY_NPM_PREFIX}/bin`,
'/usr/local/sbin',
'/usr/local/bin',
'/usr/sbin',
'/usr/bin',
'/sbin',
'/bin',
].join(':')
export type GatewayContainerSpec = {
image: string
hostPort: number
hostHome: string
envFilePath: string
gatewayToken?: string
timezone: string
}
export interface ContainerRuntimeConfig {
vm: VmRuntime
shell: ContainerCli
loader: { ensureImageLoaded(ref: string, onLog?: LogFn): Promise<void> }
projectDir: string
}
export class ContainerRuntime {
constructor(
private podman: PodmanRuntime,
private projectDir: string,
) {}
private readonly vm: VmRuntime
private readonly shell: ContainerCli
private readonly loader: {
ensureImageLoaded(ref: string, onLog?: LogFn): Promise<void>
}
private readonly projectDir: string
constructor(config: ContainerRuntimeConfig) {
this.vm = config.vm
this.shell = config.shell
this.loader = config.loader
this.projectDir = config.projectDir
}
async ensureReady(onLog?: LogFn): Promise<void> {
logger.info('Ensuring Podman runtime readiness')
return this.podman.ensureReady(onLog)
logger.info('Ensuring BrowserOS VM runtime readiness')
await this.vm.ensureReady(onLog)
await this.vm.getDefaultGateway()
}
async isPodmanAvailable(): Promise<boolean> {
return this.podman.isPodmanAvailable()
return true
}
async getMachineStatus(): Promise<{
initialized: boolean
running: boolean
}> {
return this.podman.getMachineStatus()
const running = await this.vm.isReady()
return { initialized: running, running }
}
async composeUp(onLog?: LogFn): Promise<void> {
const code = await this.compose(['up', '-d'], onLog)
if (code !== 0) throw new Error(`compose up failed with code ${code}`)
async pullImage(image: string, onLog?: LogFn): Promise<void> {
await this.loader.ensureImageLoaded(image, onLog)
}
async composeDown(onLog?: LogFn): Promise<void> {
const code = await this.compose(['down'], onLog)
if (code !== 0) throw new Error(`compose down failed with code ${code}`)
async startGateway(
input: GatewayContainerSpec,
onLog?: LogFn,
): Promise<void> {
await this.removeGatewayContainer(onLog)
await this.loader.ensureImageLoaded(input.image, onLog)
const container = await this.buildGatewayContainerSpec(input)
await this.shell.createContainer(container, onLog)
await this.shell.startContainer(container.name)
}
async composeStop(onLog?: LogFn): Promise<void> {
const code = await this.compose(['stop'], onLog)
if (code !== 0) throw new Error(`compose stop failed with code ${code}`)
async stopGateway(onLog?: LogFn): Promise<void> {
await this.removeGatewayContainer(onLog)
}
async composeRestart(onLog?: LogFn): Promise<void> {
const code = await this.compose(['restart'], onLog)
if (code !== 0) throw new Error(`compose restart failed with code ${code}`)
async restartGateway(
input: GatewayContainerSpec,
onLog?: LogFn,
): Promise<void> {
await this.startGateway(input, onLog)
}
async composePull(onLog?: LogFn): Promise<void> {
const code = await this.compose(['pull', '--quiet'], onLog)
if (code !== 0) throw new Error(`compose pull failed with code ${code}`)
}
async composeLogs(tail = 50): Promise<string[]> {
async getGatewayLogs(tail = 50): Promise<string[]> {
const lines: string[] = []
await this.compose(['logs', '--no-color', '--tail', String(tail)], (line) =>
lines.push(line),
await this.shell.runCommand(
['logs', '-n', String(tail), OPENCLAW_GATEWAY_CONTAINER_NAME],
(line) => lines.push(line),
)
return lines
}
async isHealthy(port: number): Promise<boolean> {
async isHealthy(hostPort: number): Promise<boolean> {
try {
const res = await fetch(`http://127.0.0.1:${port}/healthz`)
const res = await fetch(`http://127.0.0.1:${hostPort}/healthz`)
return res.ok
} catch {
return false
}
}
async isReady(port: number): Promise<boolean> {
async isReady(hostPort: number): Promise<boolean> {
try {
const res = await fetch(`http://127.0.0.1:${port}/readyz`)
const res = await fetch(`http://127.0.0.1:${hostPort}/readyz`)
return res.ok
} catch {
return false
}
}
async waitForReady(port: number, timeoutMs = 30_000): Promise<boolean> {
logger.info('Waiting for OpenClaw gateway readiness', { port, timeoutMs })
async waitForReady(hostPort: number, timeoutMs = 30_000): Promise<boolean> {
logger.info('Waiting for OpenClaw gateway readiness', {
hostPort,
timeoutMs,
})
const start = Date.now()
while (Date.now() - start < timeoutMs) {
if (await this.isReady(port)) {
logger.info('OpenClaw gateway became ready', {
port,
waitMs: Date.now() - start,
})
return true
}
if (await this.isReady(hostPort)) return true
await Bun.sleep(1000)
}
logger.error('Timed out waiting for OpenClaw gateway readiness', {
port,
hostPort,
timeoutMs,
})
return false
}
async copyComposeFile(sourceTemplatePath: string): Promise<void> {
await copyFile(sourceTemplatePath, join(this.projectDir, COMPOSE_FILE_NAME))
}
async writeEnvFile(content: string): Promise<void> {
await writeFile(join(this.projectDir, ENV_FILE_NAME), content, {
mode: 0o600,
})
}
/**
* Stops the Podman machine only if no non-BrowserOS containers are running.
* Prevents killing the user's own Podman workloads.
*/
async stopMachineIfSafe(): Promise<void> {
const status = await this.podman.getMachineStatus()
if (!status.running) return
try {
const containers = await this.podman.listRunningContainers()
const allOurs = containers.every((name) =>
name.startsWith(OPENCLAW_COMPOSE_PROJECT_NAME),
)
if (containers.length === 0 || allOurs) {
await this.podman.stopMachine()
}
} catch {
// Best effort — don't stop machine if we can't check
}
async stopVm(): Promise<void> {
await this.vm.stopVm()
}
async execInContainer(command: string[], onLog?: LogFn): Promise<number> {
return this.podman.runCommand(
['exec', OPENCLAW_GATEWAY_CONTAINER_NAME, ...command],
{
onOutput: onLog,
},
return this.shell.exec(OPENCLAW_GATEWAY_CONTAINER_NAME, command, onLog)
}
// Unlike execInContainer, this returns stdout and stderr separately
// so callers that need to parse program output (e.g. JSON status
// commands) aren't forced to untangle it from nerdctl's stderr.
async runInContainer(command: string[]): Promise<ContainerCommandResult> {
return this.shell.runCommand([
'exec',
OPENCLAW_GATEWAY_CONTAINER_NAME,
...command,
])
}
async runGatewaySetupCommand(
command: string[],
spec: GatewayContainerSpec,
onLog?: LogFn,
): Promise<number> {
const setupContainerName = `${OPENCLAW_GATEWAY_CONTAINER_NAME}-setup`
await this.shell.removeContainer(setupContainerName, { force: true }, onLog)
await this.loader.ensureImageLoaded(spec.image, onLog)
const setupArgs = command[0] === 'node' ? command.slice(1) : command
const createResult = await this.shell.runCommand(
[
'create',
'--name',
setupContainerName,
...(await this.buildGatewayRunArgs(spec)),
spec.image,
'node',
...setupArgs,
],
onLog,
)
if (createResult.exitCode !== 0) {
await this.shell.removeContainer(
setupContainerName,
{ force: true },
onLog,
)
return createResult.exitCode
}
try {
const startResult = await this.shell.runCommand(
['start', '-a', setupContainerName],
onLog,
)
return startResult.exitCode
} finally {
await this.shell.removeContainer(
setupContainerName,
{ force: true },
onLog,
)
}
}
tailGatewayLogs(onLine: LogFn): () => void {
return this.podman.tailContainerLogs(
return this.shell.tailLogs(OPENCLAW_GATEWAY_CONTAINER_NAME, onLine)
}
private async removeGatewayContainer(onLog?: LogFn): Promise<void> {
await this.shell.removeContainer(
OPENCLAW_GATEWAY_CONTAINER_NAME,
onLine,
{ force: true },
onLog,
)
}
private async compose(args: string[], onLog?: LogFn): Promise<number> {
const lines: string[] = []
const command = ['podman', 'compose', ...args].join(' ')
logger.info('Running OpenClaw compose command', {
command,
})
const code = await this.podman.runCommand(['compose', ...args], {
cwd: this.projectDir,
env: { COMPOSE_PROJECT_NAME: OPENCLAW_COMPOSE_PROJECT_NAME },
onOutput: (line) => {
lines.push(line)
onLog?.(line)
private async buildGatewayContainerSpec(
input: GatewayContainerSpec,
): Promise<ContainerSpec> {
return {
name: OPENCLAW_GATEWAY_CONTAINER_NAME,
image: input.image,
restart: 'unless-stopped',
ports: [
{
hostIp: '127.0.0.1',
hostPort: input.hostPort,
containerPort: OPENCLAW_GATEWAY_CONTAINER_PORT,
},
],
envFile: this.translateHostPath(input.envFilePath, input.hostHome),
env: this.buildGatewayEnv(input),
mounts: [{ source: GUEST_OPENCLAW_HOME, target: GATEWAY_CONTAINER_HOME }],
addHosts: [await this.hostContainersInternalEntry()],
health: {
cmd: `curl -sf http://127.0.0.1:${OPENCLAW_GATEWAY_CONTAINER_PORT}/healthz`,
interval: '30s',
timeout: '10s',
retries: 3,
},
})
if (code !== 0) {
logger.error('OpenClaw compose command failed', {
command,
exitCode: code,
output: lines,
})
} else {
logger.info('OpenClaw compose command succeeded', {
command,
})
command: [
'node',
'dist/index.js',
'gateway',
'--bind',
'lan',
'--port',
String(OPENCLAW_GATEWAY_CONTAINER_PORT),
'--allow-unconfigured',
],
}
}
return code
private async buildGatewayRunArgs(
input: GatewayContainerSpec,
): Promise<string[]> {
const args = [
'--env-file',
this.translateHostPath(input.envFilePath, input.hostHome),
'-v',
`${GUEST_OPENCLAW_HOME}:${GATEWAY_CONTAINER_HOME}`,
]
for (const [key, value] of Object.entries(this.buildGatewayEnv(input))) {
args.push('-e', `${key}=${value}`)
}
args.push('--add-host', await this.hostContainersInternalEntry())
return args
}
private async hostContainersInternalEntry(): Promise<string> {
return `host.containers.internal:${await this.vm.getDefaultGateway()}`
}
private buildGatewayEnv(input: GatewayContainerSpec): Record<string, string> {
return {
HOME: GATEWAY_CONTAINER_HOME,
OPENCLAW_HOME: GATEWAY_CONTAINER_HOME,
OPENCLAW_STATE_DIR: GATEWAY_STATE_DIR,
OPENCLAW_NO_RESPAWN: '1',
NODE_COMPILE_CACHE: '/var/tmp/openclaw-compile-cache',
NODE_ENV: 'production',
TZ: input.timezone,
PATH: GATEWAY_PATH,
NPM_CONFIG_PREFIX: GATEWAY_NPM_PREFIX,
...(input.gatewayToken
? { OPENCLAW_GATEWAY_TOKEN: input.gatewayToken }
: {}),
}
}
private translateHostPath(path: string, openclawHostDir: string): string {
if (path === openclawHostDir) return GUEST_OPENCLAW_HOME
if (path.startsWith(`${openclawHostDir}/`)) {
return `${GUEST_OPENCLAW_HOME}${path.slice(openclawHostDir.length)}`
}
return hostPathToGuest(path)
}
}

View File

@@ -27,3 +27,10 @@ export class OpenClawProtectedAgentError extends Error {
this.name = 'OpenClawProtectedAgentError'
}
}
export class OpenClawSessionNotFoundError extends Error {
constructor(public readonly sessionKey: string) {
super(`OpenClaw session not found: ${sessionKey}`)
this.name = 'OpenClawSessionNotFoundError'
}
}

View File

@@ -1,754 +0,0 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* WebSocket client for the OpenClaw Gateway protocol.
* Handles handshake (challenge → connect → hello-ok) with Ed25519 device
* identity signing, JSON-RPC over WS, and auto-reconnect.
* Used for agent CRUD and health — chat uses HTTP.
*/
import crypto from 'node:crypto'
import { mkdirSync, readFileSync, writeFileSync } from 'node:fs'
import { join } from 'node:path'
import { OPENCLAW_CONTAINER_HOME } from '@browseros/shared/constants/openclaw'
import { logger } from '../../../lib/logger'
const RPC_TIMEOUT_MS = 15_000
const SCOPES = [
'operator.read',
'operator.write',
'operator.admin',
'operator.approvals',
'operator.pairing',
]
interface DeviceIdentity {
deviceId: string
publicKeyPem: string
privateKeyPem: string
}
interface PendingRequest {
resolve: (value: unknown) => void
reject: (reason: Error) => void
timer: ReturnType<typeof setTimeout>
}
interface WsFrame {
type: 'req' | 'res' | 'event'
id?: string
method?: string
params?: Record<string, unknown>
ok?: boolean
payload?: Record<string, unknown>
error?: { message: string; code?: string }
event?: string
}
export type GatewayClientConnectionState =
| 'idle'
| 'connecting'
| 'connected'
| 'closed'
| 'failed'
export interface GatewayHandshakeError {
code?: string
message: string
}
export interface OpenClawStreamEvent {
type:
| 'text-delta'
| 'thinking'
| 'tool-start'
| 'tool-end'
| 'tool-output'
| 'lifecycle'
| 'done'
| 'error'
data: Record<string, unknown>
}
export interface GatewayAgentEntry {
agentId: string
name: string
workspace: string
model?: string
}
// ── Device Identity Helpers ─────────────────────────────────────────
function rawPublicKeyFromPem(pem: string): Buffer {
const der = Buffer.from(
pem.replace(/-----[^-]+-----/g, '').replace(/\s/g, ''),
'base64',
)
return der.subarray(12)
}
function signChallenge(
device: DeviceIdentity,
nonce: string,
token: string,
): { signature: string; signedAt: number; publicKey: string } {
const signedAt = Date.now()
const payload = `v3|${device.deviceId}|cli|cli|operator|${SCOPES.join(',')}|${signedAt}|${token}|${nonce}|${process.platform}|`
const privateKey = crypto.createPrivateKey(device.privateKeyPem)
const sig = crypto.sign(null, Buffer.from(payload, 'utf-8'), privateKey)
return {
signature: sig.toString('base64url'),
signedAt,
publicKey: rawPublicKeyFromPem(device.publicKeyPem).toString('base64url'),
}
}
/**
* Generates a client Ed25519 identity and pre-seeds it into the gateway's
* paired devices file so the gateway trusts it on next boot.
* Must be called before compose up (or requires a restart after).
*/
export function ensureClientIdentity(openclawDir: string): DeviceIdentity {
const identityPath = join(openclawDir, 'client-identity.json')
try {
return JSON.parse(readFileSync(identityPath, 'utf-8'))
} catch {
// Generate new identity
}
const { publicKey, privateKey } = crypto.generateKeyPairSync('ed25519')
const publicKeyPem = publicKey
.export({ type: 'spki', format: 'pem' })
.toString()
const privateKeyPem = privateKey
.export({ type: 'pkcs8', format: 'pem' })
.toString()
const rawPub = rawPublicKeyFromPem(publicKeyPem)
const deviceId = crypto.createHash('sha256').update(rawPub).digest('hex')
const identity: DeviceIdentity = { deviceId, publicKeyPem, privateKeyPem }
writeFileSync(identityPath, JSON.stringify(identity, null, 2), {
mode: 0o600,
})
seedPairedDevice(openclawDir, identity)
logger.info('Generated client device identity and pre-seeded pairing')
return identity
}
function seedPairedDevice(openclawDir: string, identity: DeviceIdentity): void {
const devicesDir = join(openclawDir, 'devices')
mkdirSync(devicesDir, { recursive: true })
const pairedPath = join(devicesDir, 'paired.json')
let paired: Record<string, unknown> = {}
try {
paired = JSON.parse(readFileSync(pairedPath, 'utf-8'))
} catch {
// First time
}
const rawPub = rawPublicKeyFromPem(identity.publicKeyPem)
paired[identity.deviceId] = {
deviceId: identity.deviceId,
publicKey: rawPub.toString('base64url'),
platform: process.platform,
clientId: 'cli',
clientMode: 'cli',
role: 'operator',
roles: ['operator'],
scopes: SCOPES,
pairedAt: Date.now(),
label: 'browseros-server',
}
writeFileSync(pairedPath, JSON.stringify(paired, null, 2), { mode: 0o600 })
}
// ── Gateway Client ──────────────────────────────────────────────────
export class GatewayClient {
private ws: WebSocket | null = null
private _connected = false
private pendingRequests = new Map<string, PendingRequest>()
private device: DeviceIdentity | null = null
private connectionState: GatewayClientConnectionState = 'idle'
private lastHandshakeError: GatewayHandshakeError | null = null
constructor(
private readonly port: number,
private readonly token: string,
private readonly openclawDir: string,
private readonly version = '1.0.0',
) {
try {
const identityPath = join(this.openclawDir, 'client-identity.json')
this.device = JSON.parse(readFileSync(identityPath, 'utf-8'))
} catch {
logger.warn('Client device identity not found, WS auth may fail')
}
}
get isConnected(): boolean {
return this._connected
}
get state(): GatewayClientConnectionState {
return this.connectionState
}
get lastError(): GatewayHandshakeError | null {
return this.lastHandshakeError
}
async connect(): Promise<void> {
return new Promise((resolve, reject) => {
this.connectionState = 'connecting'
this.lastHandshakeError = null
logger.info('Connecting to OpenClaw Gateway WS', {
port: this.port,
hasDeviceIdentity: !!this.device,
})
this.ws = new WebSocket(`ws://127.0.0.1:${this.port}`, {
headers: { Origin: `http://127.0.0.1:${this.port}` },
} as unknown as string[])
let handshakeComplete = false
let connectReqId: string | null = null
this.ws.onmessage = (event) => {
const frame = GatewayClient.parseFrame(event.data)
if (!frame) return
if (!handshakeComplete) {
if (frame.type === 'event' && frame.event === 'connect.challenge') {
const nonce = (frame.payload as Record<string, unknown>)
?.nonce as string
logger.info('Received OpenClaw Gateway challenge', {
hasNonce: !!nonce,
hasDeviceIdentity: !!this.device,
})
connectReqId = globalThis.crypto.randomUUID()
const params: Record<string, unknown> = {
minProtocol: 3,
maxProtocol: 3,
client: {
id: 'cli',
version: this.version,
platform: process.platform,
mode: 'cli',
},
role: 'operator',
scopes: SCOPES,
caps: [],
commands: [],
permissions: {},
auth: { token: this.token },
locale: 'en-US',
userAgent: `browseros-server/${this.version}`,
}
if (this.device && nonce) {
const signed = signChallenge(this.device, nonce, this.token)
params.device = {
id: this.device.deviceId,
publicKey: signed.publicKey,
signature: signed.signature,
signedAt: signed.signedAt,
nonce,
}
}
this.ws?.send(
JSON.stringify({
type: 'req',
id: connectReqId,
method: 'connect',
params,
}),
)
return
}
if (frame.type === 'res' && frame.id === connectReqId) {
if (frame.ok) {
handshakeComplete = true
this._connected = true
this.connectionState = 'connected'
logger.info('Gateway WS connected')
resolve()
} else {
const msg = frame.error?.message ?? 'Handshake failed'
this.connectionState = 'failed'
this.lastHandshakeError = {
message: msg,
code: frame.error?.code,
}
logger.error('Gateway WS handshake rejected', {
error: msg,
code: frame.error?.code,
})
reject(new Error(msg))
}
return
}
return
}
this.resolvePendingRequest(frame)
}
this.ws.onerror = (err) => {
logger.error('Gateway WS socket error', {
error: err instanceof Error ? err.message : 'unknown',
handshakeComplete,
})
if (!handshakeComplete) {
this.connectionState = 'failed'
reject(
new Error(
`WS connection error: ${err instanceof Error ? err.message : 'unknown'}`,
),
)
}
}
this.ws.onclose = () => {
this._connected = false
this.connectionState = 'closed'
this.rejectAllPending('WebSocket closed')
if (handshakeComplete) {
logger.info('Gateway WS disconnected')
}
this.ws = null
}
})
}
disconnect(): void {
this._connected = false
this.connectionState = 'closed'
this.rejectAllPending('Client disconnecting')
if (this.ws) {
this.ws.onclose = null
this.ws.close()
this.ws = null
}
}
// ── RPC ──────────────────────────────────────────────────────────────
async rpc<T = Record<string, unknown>>(
method: string,
params: Record<string, unknown> = {},
): Promise<T> {
if (!this._connected || !this.ws) {
throw new Error('Gateway WS not connected')
}
const id = globalThis.crypto.randomUUID()
return new Promise<T>((resolve, reject) => {
const timer = setTimeout(() => {
this.pendingRequests.delete(id)
reject(new Error(`RPC timeout: ${method}`))
}, RPC_TIMEOUT_MS)
this.pendingRequests.set(id, {
resolve: resolve as (value: unknown) => void,
reject,
timer,
})
this.ws?.send(JSON.stringify({ type: 'req', id, method, params }))
})
}
// ── Agent Methods ────────────────────────────────────────────────────
async listAgents(): Promise<GatewayAgentEntry[]> {
const result = await this.rpc<{
agents: Array<{
id: string
name?: string
workspace: string
model?: string
}>
}>('agents.list')
return (result.agents ?? []).map((a) => ({
agentId: a.id,
name: a.name ?? a.id,
workspace: a.workspace,
model: a.model,
}))
}
async createAgent(input: {
name: string
workspace: string
model?: string
}): Promise<GatewayAgentEntry> {
const result = await this.rpc<{
agentId?: string
id?: string
name?: string
workspace?: string
model?: string
}>('agents.create', input)
return {
agentId: result.agentId ?? result.id ?? input.name,
name: result.name ?? input.name,
workspace: result.workspace ?? input.workspace,
model: result.model ?? input.model,
}
}
async deleteAgent(agentId: string): Promise<void> {
await this.rpc('agents.delete', { id: agentId })
}
// ── Health ───────────────────────────────────────────────────────────
async getHealth(): Promise<Record<string, unknown>> {
return this.rpc('health')
}
// ── Chat Stream ─────────────────────────────────────────────────────
chatStream(
agentId: string,
sessionKey: string,
message: string,
): ReadableStream<OpenClawStreamEvent> {
if (!this._connected) {
throw new Error('Gateway WS not connected')
}
const fullSessionKey = `agent:${agentId}:browseros-${sessionKey}`
const idempotencyKey = globalThis.crypto.randomUUID()
const streamClient = new GatewayClient(
this.port,
this.token,
this.openclawDir,
this.version,
)
return new ReadableStream<OpenClawStreamEvent>({
start: async (controller) => {
try {
await streamClient.connect()
} catch (error) {
controller.enqueue({
type: 'error',
data: {
message:
error instanceof Error
? error.message
: 'Gateway WS not connected',
},
})
controller.close()
return
}
const ws = streamClient.ws
if (!ws) {
controller.enqueue({
type: 'error',
data: { message: 'Gateway WS not connected' },
})
controller.close()
return
}
const subscribeId = globalThis.crypto.randomUUID()
const agentReqId = globalThis.crypto.randomUUID()
let finished = false
const finish = (event?: OpenClawStreamEvent) => {
if (finished) return
finished = true
if (event) controller.enqueue(event)
controller.close()
streamClient.disconnect()
}
ws.onmessage = (event) => {
const frame = GatewayClient.parseFrame(event.data)
if (!frame) return
if (
this.handleChatStreamControlFrame(
frame,
subscribeId,
agentReqId,
finish,
)
) {
return
}
this.handleChatStreamEventFrame(frame, controller, finish)
}
ws.onclose = () => {
if (finished) return
finish({
type: 'error',
data: { message: 'Gateway WS disconnected' },
})
}
ws.onerror = () => {
if (finished) return
finish({
type: 'error',
data: { message: 'Gateway WS connection error' },
})
}
ws.send(
JSON.stringify({
type: 'req',
id: subscribeId,
method: 'sessions.subscribe',
params: { sessionKey: fullSessionKey },
}),
)
ws.send(
JSON.stringify({
type: 'req',
id: agentReqId,
method: 'agent',
params: {
message,
sessionKey: fullSessionKey,
idempotencyKey,
},
}),
)
},
cancel: () => {
if (streamClient.ws?.readyState === WebSocket.OPEN) {
streamClient.ws.send(
JSON.stringify({
type: 'req',
id: globalThis.crypto.randomUUID(),
method: 'sessions.abort',
params: { sessionKey: fullSessionKey },
}),
)
}
streamClient.disconnect()
},
})
}
// ── Helpers ──────────────────────────────────────────────────────────
static agentWorkspace(name: string): string {
return name === 'main'
? `${OPENCLAW_CONTAINER_HOME}/workspace`
: `${OPENCLAW_CONTAINER_HOME}/workspace-${name}`
}
private static parseFrame(data: unknown): WsFrame | null {
try {
return JSON.parse(
typeof data === 'string'
? data
: new TextDecoder().decode(data as ArrayBuffer),
) as WsFrame
} catch {
return null
}
}
private rejectAllPending(reason: string): void {
for (const [id, pending] of this.pendingRequests) {
clearTimeout(pending.timer)
pending.reject(new Error(reason))
this.pendingRequests.delete(id)
}
}
private resolvePendingRequest(frame: WsFrame): void {
if (frame.type !== 'res' || !frame.id) return
const pending = this.pendingRequests.get(frame.id)
if (!pending) return
this.pendingRequests.delete(frame.id)
clearTimeout(pending.timer)
if (frame.ok) {
pending.resolve(frame.payload)
} else {
pending.reject(new Error(frame.error?.message ?? 'RPC error'))
}
}
private handleChatStreamControlFrame(
frame: WsFrame,
subscribeId: string,
agentReqId: string,
finish: (event?: OpenClawStreamEvent) => void,
): boolean {
if (frame.type !== 'res' || !frame.id) return false
if (frame.id !== subscribeId && frame.id !== agentReqId) return false
if (!frame.ok) {
finish({
type: 'error',
data: {
message: frame.error?.message ?? 'RPC error',
code: frame.error?.code,
},
})
}
return true
}
private handleChatStreamEventFrame(
frame: WsFrame,
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
finish: (event?: OpenClawStreamEvent) => void,
): void {
if (frame.type !== 'event' || !frame.event || !frame.payload) return
switch (frame.event) {
case 'agent':
this.handleAgentStreamEvent(frame.payload, controller)
return
case 'session.tool':
this.handleSessionToolStreamEvent(frame.payload, controller)
return
case 'session.message':
this.handleSessionMessageStreamEvent(frame.payload, controller)
return
case 'chat':
this.handleChatCompletionEvent(frame.payload, finish)
return
default:
return
}
}
private handleAgentStreamEvent(
payload: Record<string, unknown>,
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
): void {
const streamType = payload.stream as string | undefined
const data = payload.data as Record<string, unknown> | undefined
if (streamType === 'assistant' && data?.delta) {
controller.enqueue({
type: 'text-delta',
data: { text: data.delta },
})
return
}
if (streamType === 'item' && data) {
const phase = data.phase as string | undefined
if (phase === 'start') {
controller.enqueue({
type: 'tool-start',
data: {
toolCallId: data.toolCallId ?? data.id,
toolName: data.name ?? data.title,
kind: data.kind,
},
})
return
}
if (phase === 'end') {
controller.enqueue({
type: 'tool-end',
data: {
toolCallId: data.toolCallId ?? data.id,
status: data.status,
durationMs: data.durationMs,
},
})
return
}
}
if (streamType === 'lifecycle') {
controller.enqueue({
type: 'lifecycle',
data: { phase: data?.phase ?? payload.phase },
})
}
}
private handleSessionToolStreamEvent(
payload: Record<string, unknown>,
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
): void {
const toolData = (payload.data as Record<string, unknown>) ?? payload
const phase = (toolData.phase as string) ?? (payload.phase as string)
if (phase !== 'result') return
controller.enqueue({
type: 'tool-output',
data: {
toolCallId: toolData.toolCallId,
isError: toolData.isError ?? false,
meta: toolData.meta,
},
})
}
private handleSessionMessageStreamEvent(
payload: Record<string, unknown>,
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
): void {
const message = payload.message as Record<string, unknown> | undefined
if (message?.role !== 'assistant') return
const content = message.content as
| Array<Record<string, unknown>>
| undefined
if (!content) return
for (const block of content) {
if (block.type !== 'thinking') continue
const text =
(block.thinking as string) ??
(block.content as string) ??
(block.text as string) ??
''
if (!text) continue
controller.enqueue({
type: 'thinking',
data: { text },
})
}
}
private handleChatCompletionEvent(
payload: Record<string, unknown>,
finish: (event?: OpenClawStreamEvent) => void,
): void {
if ((payload.state as string | undefined) !== 'final') return
finish({
type: 'done',
data: { text: (payload.text as string) ?? '' },
})
}
}

View File

@@ -0,0 +1,581 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { OPENCLAW_CONTAINER_HOME } from '@browseros/shared/constants/openclaw'
type LogFn = (line: string) => void
interface ContainerExecutor {
execInContainer(command: string[], onLog?: LogFn): Promise<number>
}
export interface OpenClawConfigBatchEntry {
path: string
value: unknown
}
interface RawAgentRecord {
id: string
name?: string
workspace: string
model?: string
}
export interface OpenClawAgentRecord {
agentId: string
name: string
workspace: string
model?: string
}
export interface OpenClawSessionEntry {
key: string
updatedAt: number
sessionId: string
agentId: string
kind: string
status?: string
totalTokens?: number
model?: string
modelProvider?: string
}
export interface OpenClawChatBlock {
type: 'text' | 'toolCall' | 'thinking'
text?: string
name?: string
arguments?: unknown
thinking?: string
}
export interface OpenClawChatMessage {
role: 'user' | 'assistant' | 'toolResult'
content: OpenClawChatBlock[]
timestamp?: number
usage?: { input: number; output: number }
stopReason?: string
toolName?: string
toolCallId?: string
isError?: boolean
}
export class OpenClawCliClient {
constructor(private readonly executor: ContainerExecutor) {}
async runOnboard(
input: {
acceptRisk?: boolean
authChoice?: string
customBaseUrl?: string
customCompatibility?: 'anthropic' | 'openai-completions'
customModelId?: string
customProviderId?: string
gatewayAuth?: 'none' | 'password' | 'token'
gatewayBind?: 'auto' | 'custom' | 'lan' | 'loopback' | 'tailnet'
gatewayPort?: number
gatewayToken?: string
gatewayTokenRefEnv?: string
installDaemon?: boolean
mode?: 'local' | 'remote'
nonInteractive?: boolean
reset?: boolean
resetScope?: 'config' | 'config+creds+sessions' | 'full'
secretInputMode?: 'plain' | 'ref'
skipHealth?: boolean
workspace?: string
} = {},
): Promise<void> {
const args = ['onboard']
if (input.nonInteractive) {
args.push('--non-interactive')
}
if (input.mode) {
args.push('--mode', input.mode)
}
if (input.workspace) {
args.push('--workspace', input.workspace)
}
if (input.reset) {
args.push('--reset')
}
if (input.resetScope) {
args.push('--reset-scope', input.resetScope)
}
if (input.authChoice) {
args.push('--auth-choice', input.authChoice)
}
if (input.secretInputMode) {
args.push('--secret-input-mode', input.secretInputMode)
}
if (input.customBaseUrl) {
args.push('--custom-base-url', input.customBaseUrl)
}
if (input.customModelId) {
args.push('--custom-model-id', input.customModelId)
}
if (input.customProviderId) {
args.push('--custom-provider-id', input.customProviderId)
}
if (input.customCompatibility) {
args.push('--custom-compatibility', input.customCompatibility)
}
if (input.gatewayAuth) {
args.push('--gateway-auth', input.gatewayAuth)
}
if (input.gatewayToken) {
args.push('--gateway-token', input.gatewayToken)
}
if (input.gatewayTokenRefEnv) {
args.push('--gateway-token-ref-env', input.gatewayTokenRefEnv)
}
if (input.gatewayPort) {
args.push('--gateway-port', String(input.gatewayPort))
}
if (input.gatewayBind) {
args.push('--gateway-bind', input.gatewayBind)
}
if (input.installDaemon === true) {
args.push('--install-daemon')
} else if (input.installDaemon === false) {
args.push('--no-install-daemon')
}
if (input.skipHealth) {
args.push('--skip-health')
}
if (input.acceptRisk) {
args.push('--accept-risk')
}
await this.runCommand(args)
}
async setConfig(path: string, value: unknown): Promise<void> {
await this.runCommand(['config', 'set', path, formatConfigValue(value)])
}
async setConfigBatch(entries: OpenClawConfigBatchEntry[]): Promise<void> {
await this.runCommand([
'config',
'set',
'--batch-json',
JSON.stringify(entries),
])
}
async getConfig(path: string): Promise<unknown> {
const output = await this.runCommand(['config', 'get', path])
return parseConfigValue(output)
}
async validateConfig(): Promise<unknown> {
const output = await this.runCommand(['config', 'validate', '--json'])
return parseConfigValue(output)
}
async setDefaultModel(model: string): Promise<void> {
await this.runCommand(['models', 'set', model])
}
async listAgents(): Promise<OpenClawAgentRecord[]> {
const records = await this.runAgentListCommand()
const agents = Array.isArray(records) ? records : (records.agents ?? [])
return agents.map((record) => ({
agentId: record.id,
name: record.name ?? record.id,
workspace: record.workspace,
model: record.model,
}))
}
async createAgent(input: {
name: string
model?: string
}): Promise<OpenClawAgentRecord> {
const workspace = this.agentWorkspace(input.name)
const args = ['agents', 'add', input.name, '--workspace', workspace]
if (input.model) {
args.push('--model', input.model)
}
args.push('--non-interactive', '--json')
await this.runCommand(args)
const agents = await this.listAgents()
const agent = agents.find((entry) => entry.agentId === input.name)
if (!agent) {
throw new Error(`Created agent ${input.name} was not found in agent list`)
}
return agent
}
async deleteAgent(agentId: string): Promise<void> {
await this.runCommand(['agents', 'delete', agentId, '--force', '--json'])
}
async probe(): Promise<void> {
await this.listAgents()
}
async listSessions(agentId?: string): Promise<OpenClawSessionEntry[]> {
const args = ['sessions', '--json']
if (agentId) {
args.push('--agent', agentId)
} else {
args.push('--all-agents')
}
const output = await this.runCommand(args)
const parsed = parseFirstMatchingJson<
{ sessions?: unknown[]; count?: number } | unknown[]
>(output, isSessionListPayload)
if (parsed === null) {
throw new Error(
`Failed to parse OpenClaw sessions output: ${output.slice(0, 200)}`,
)
}
const entries = Array.isArray(parsed) ? parsed : (parsed.sessions ?? [])
return entries.map(toSessionEntry)
}
async getChatHistory(sessionKey: string): Promise<OpenClawChatMessage[]> {
const output = await this.runCommand([
'gateway',
'call',
'chat.history',
'--params',
JSON.stringify({ sessionKey }),
'--json',
])
const parsed = parseFirstMatchingJson<{ messages?: unknown[] }>(
output,
(value) => isPlainObject(value) && 'messages' in value,
)
if (parsed === null) {
throw new Error(
`Failed to parse OpenClaw chat history output: ${output.slice(0, 200)}`,
)
}
return (parsed.messages ?? []).map(toChatMessage)
}
private agentWorkspace(name: string): string {
return name === 'main'
? `${OPENCLAW_CONTAINER_HOME}/workspace`
: `${OPENCLAW_CONTAINER_HOME}/workspace-${name}`
}
private async runCommand(args: string[]): Promise<string> {
const output: string[] = []
const command = ['node', 'dist/index.js', ...args]
const exitCode = await this.executor.execInContainer(command, (line) => {
output.push(line)
})
if (exitCode !== 0) {
const detail = output.join('\n').trim()
throw new Error(
detail || `OpenClaw command failed (${args.slice(0, 2).join(' ')})`,
)
}
return output.join('\n').trim()
}
private async runAgentListCommand(): Promise<
RawAgentRecord[] | { agents?: RawAgentRecord[] }
> {
const output = await this.runCommand(['agents', 'list', '--json'])
return parseAgentListOutput(output)
}
}
function formatConfigValue(value: unknown): string {
if (typeof value === 'string') return value
return JSON.stringify(value)
}
function parseConfigValue(output: string): unknown {
const parsed = selectConfigJson<unknown>(output)
return parsed ?? output
}
function parseAgentListOutput(
output: string,
): RawAgentRecord[] | { agents?: RawAgentRecord[] } {
const parsed = parseFirstMatchingJson<
RawAgentRecord[] | { agents?: RawAgentRecord[] }
>(output, isAgentListPayload)
if (parsed !== null) return parsed
throw new Error(
`Failed to parse OpenClaw JSON output: ${output.slice(0, 200)}`,
)
}
function parseFirstMatchingJson<T>(
output: string,
predicate?: (value: unknown) => boolean,
): T | null {
const candidates = collectJsonCandidates(output)
for (const candidate of candidates) {
const parsed = tryParseJson<T>(candidate)
if (parsed === null) continue
if (predicate && !predicate(parsed)) continue
return parsed
}
return null
}
function selectConfigJson<T>(output: string): T | null {
const candidates = collectJsonCandidates(output)
const parsedCandidates: Array<{ text: string; value: T }> = []
for (const candidate of candidates) {
const parsed = tryParseJson<T>(candidate)
if (parsed === null) continue
if (isStructuredLogPayload(parsed)) continue
parsedCandidates.push({ text: candidate, value: parsed })
}
if (parsedCandidates.length === 0) return null
return parsedCandidates.reduce((best, candidate) =>
candidate.text.length > best.text.length ? candidate : best,
).value
}
function collectJsonCandidates(output: string): string[] {
const candidates = [output.trim()]
for (const line of output.split(/\r?\n/)) {
const trimmed = line.trim()
if (trimmed) candidates.push(trimmed)
}
for (let index = 0; index < output.length; index += 1) {
const char = output[index]
if (char !== '[' && char !== '{') continue
const extracted = extractJsonSubstring(output, index)
if (extracted) {
candidates.push(extracted)
}
}
return candidates
}
function extractJsonSubstring(
output: string,
startIndex: number,
): string | null {
const opening = output[startIndex]
const closing = opening === '{' ? '}' : ']'
const stack: string[] = [closing]
let inString = false
let escaped = false
for (let index = startIndex + 1; index < output.length; index += 1) {
const char = output[index]
if (inString) {
if (escaped) {
escaped = false
continue
}
if (char === '\\') {
escaped = true
continue
}
if (char === '"') {
inString = false
}
continue
}
if (char === '"') {
inString = true
continue
}
if (char === '{') {
stack.push('}')
continue
}
if (char === '[') {
stack.push(']')
continue
}
const expectedClosing = stack[stack.length - 1]
if (char === expectedClosing) {
stack.pop()
if (stack.length === 0) {
return output.slice(startIndex, index + 1)
}
}
}
return null
}
function tryParseJson<T>(value: string): T | null {
const trimmed = value.trim()
if (!trimmed) return null
try {
return JSON.parse(trimmed) as T
} catch {
return null
}
}
function isAgentListPayload(
value: unknown,
): value is RawAgentRecord[] | { agents?: RawAgentRecord[] } {
if (Array.isArray(value)) {
return value.every(isRawAgentRecord)
}
if (!isPlainObject(value)) return false
if (!('agents' in value)) return false
const agents = (value as { agents?: unknown }).agents
return (
agents === undefined ||
(Array.isArray(agents) && agents.every(isRawAgentRecord))
)
}
function isRawAgentRecord(value: unknown): value is RawAgentRecord {
return (
isPlainObject(value) &&
typeof value.id === 'string' &&
typeof value.workspace === 'string' &&
(value.name === undefined || typeof value.name === 'string') &&
(value.model === undefined || typeof value.model === 'string')
)
}
function isPlainObject(value: unknown): value is Record<string, unknown> {
return typeof value === 'object' && value !== null && !Array.isArray(value)
}
function isStructuredLogPayload(value: unknown): boolean {
if (!isPlainObject(value)) return false
return (
typeof value.level === 'string' &&
(typeof value.message === 'string' || typeof value.msg === 'string')
)
}
function isSessionListPayload(value: unknown): boolean {
if (Array.isArray(value)) return true
if (!isPlainObject(value)) return false
return 'sessions' in value || 'count' in value
}
function toSessionEntry(raw: unknown): OpenClawSessionEntry {
const record = isPlainObject(raw) ? raw : {}
return {
key: String(record.key ?? ''),
updatedAt: typeof record.updatedAt === 'number' ? record.updatedAt : 0,
sessionId: String(record.sessionId ?? ''),
agentId: String(record.agentId ?? ''),
kind: String(record.kind ?? ''),
status: typeof record.status === 'string' ? record.status : undefined,
totalTokens:
typeof record.totalTokens === 'number' ? record.totalTokens : undefined,
model: typeof record.model === 'string' ? record.model : undefined,
modelProvider:
typeof record.modelProvider === 'string'
? record.modelProvider
: undefined,
}
}
function toChatMessage(raw: unknown): OpenClawChatMessage {
const record = isPlainObject(raw) ? raw : {}
const role = isOpenClawMessageRole(record.role) ? record.role : 'assistant'
const message: OpenClawChatMessage = {
role,
content: toChatBlocks(record.content),
}
if (typeof record.timestamp === 'number') message.timestamp = record.timestamp
if (isPlainObject(record.usage)) {
const { input, output } = record.usage
if (typeof input === 'number' && typeof output === 'number') {
message.usage = { input, output }
}
}
if (typeof record.stopReason === 'string') {
message.stopReason = record.stopReason
}
if (typeof record.toolName === 'string') message.toolName = record.toolName
if (typeof record.toolCallId === 'string') {
message.toolCallId = record.toolCallId
}
if (typeof record.isError === 'boolean') message.isError = record.isError
return message
}
function toChatBlocks(content: unknown): OpenClawChatBlock[] {
if (typeof content === 'string') {
return [{ type: 'text', text: content }]
}
if (!Array.isArray(content)) return []
const blocks: OpenClawChatBlock[] = []
for (const rawBlock of content) {
if (!isPlainObject(rawBlock)) continue
if (rawBlock.type === 'toolCall') {
const block: OpenClawChatBlock = { type: 'toolCall' }
if (typeof rawBlock.name === 'string') block.name = rawBlock.name
if (rawBlock.arguments !== undefined) {
block.arguments = rawBlock.arguments
}
blocks.push(block)
continue
}
if (rawBlock.type === 'thinking') {
const block: OpenClawChatBlock = { type: 'thinking' }
if (typeof rawBlock.thinking === 'string') {
block.thinking = rawBlock.thinking
}
blocks.push(block)
continue
}
const block: OpenClawChatBlock = { type: 'text' }
if (typeof rawBlock.text === 'string') block.text = rawBlock.text
blocks.push(block)
}
return blocks
}
function isOpenClawMessageRole(
value: unknown,
): value is OpenClawChatMessage['role'] {
return value === 'user' || value === 'assistant' || value === 'toolResult'
}

View File

@@ -0,0 +1,72 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import type {
OpenClawCliProvider,
OpenClawCliProviderAuthStatus,
} from './types'
const CLAUDE_CLI_MODELS = [
'claude-sonnet-4-6',
'claude-opus-4-6',
'claude-haiku-4-5',
] as const
// `claude auth status` emits JSON on both the logged-in (exit 0) and
// not-logged-in (exit 1) paths. The caller passes us stdout alone —
// the exec layer separates stdout and stderr so no extraction or
// stripping of nerdctl noise is needed.
interface ClaudeAuthStatusPayload {
loggedIn?: boolean
email?: string
subscriptionType?: string
}
function parseClaudeAuthStatus(
stdout: string,
exitCode: number,
): OpenClawCliProviderAuthStatus {
const trimmed = stdout.trim()
// Binary missing: claude isn't installed / not on PATH.
if (exitCode === 127 || !trimmed) {
return { installed: false, loggedIn: false }
}
let payload: ClaudeAuthStatusPayload
try {
payload = JSON.parse(trimmed) as ClaudeAuthStatusPayload
} catch {
return {
installed: true,
loggedIn: false,
error: `Unexpected claude auth status output: ${trimmed.slice(0, 200)}`,
}
}
return {
installed: true,
loggedIn: !!payload.loggedIn,
accountLabel: payload.email,
subscriptionLabel: payload.subscriptionType,
}
}
export const CLAUDE_CLI_PROVIDER: OpenClawCliProvider = {
id: 'claude-cli',
displayName: 'Anthropic Claude CLI',
description: 'Uses your Claude.ai subscription via the Claude Code CLI',
npmPackage: '@anthropic-ai/claude-code',
npmPackageVersion: '2.1.119',
binary: 'claude',
authStatusCommand: ['claude', 'auth', 'status'],
// `claude auth login` in 2.1.x silently discards stdin. The REPL's
// `/login` slash command, launched from a fresh `claude` invocation,
// does accept a pasted token.
authLoginCommand: 'claude /login',
models: CLAUDE_CLI_MODELS,
parseAuthStatus: parseClaudeAuthStatus,
}

View File

@@ -0,0 +1,32 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Registry of OpenClaw CLI-backed providers. Add entries here as we
* enable more (Gemini CLI, Codex CLI, etc.).
*/
import { CLAUDE_CLI_PROVIDER } from './claude-cli'
import type { OpenClawCliProvider } from './types'
export const OPENCLAW_CLI_PROVIDERS: readonly OpenClawCliProvider[] = [
CLAUDE_CLI_PROVIDER,
]
export function getOpenClawCliProvider(
id: string,
): OpenClawCliProvider | undefined {
return OPENCLAW_CLI_PROVIDERS.find((provider) => provider.id === id)
}
export function isOpenClawCliProviderId(id: string): boolean {
return OPENCLAW_CLI_PROVIDERS.some((provider) => provider.id === id)
}
export function buildOpenClawCliProviderModelRef(
providerId: string,
modelId: string,
): string {
return `${providerId}/${modelId}`
}

View File

@@ -0,0 +1,39 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* OpenClaw CLI-backed provider registry types.
*
* A "CLI provider" is a tool that runs inside the OpenClaw gateway
* container (e.g. Claude Code CLI, Gemini CLI). OpenClaw spawns the
* binary as a subprocess when the active model is prefixed with the
* provider id — so our job is to install the tool and surface its
* auth status to the user. No Anthropic/OpenRouter-style API key.
*/
export interface OpenClawCliProviderAuthStatus {
installed: boolean
loggedIn: boolean
accountLabel?: string
subscriptionLabel?: string
error?: string
}
export interface OpenClawCliProvider {
id: string
displayName: string
description: string
npmPackage: string
// Pinned package version. npm installs go through argv directly
// (no shell), so `@latest` drift can't silently ship through.
npmPackageVersion: string
binary: string
authStatusCommand: string[]
authLoginCommand: string
models: readonly string[]
parseAuthStatus: (
stdout: string,
exitCode: number,
) => OpenClawCliProviderAuthStatus
}

View File

@@ -1,279 +0,0 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Pure functions for building OpenClaw bootstrap configuration.
* Config is write-once at setup — agent CRUD uses WS RPC, not config edits.
*/
import {
OPENCLAW_CONTAINER_HOME,
OPENCLAW_GATEWAY_PORT,
} from '@browseros/shared/constants/openclaw'
import { DEFAULT_PORTS } from '@browseros/shared/constants/ports'
const OPENCLAW_IMAGE = 'ghcr.io/openclaw/openclaw:latest'
export const PROVIDER_ENV_MAP: Record<string, string> = {
anthropic: 'ANTHROPIC_API_KEY',
openai: 'OPENAI_API_KEY',
google: 'GEMINI_API_KEY',
openrouter: 'OPENROUTER_API_KEY',
moonshot: 'MOONSHOT_API_KEY',
groq: 'GROQ_API_KEY',
mistral: 'MISTRAL_API_KEY',
}
export interface OpenClawProviderInput {
providerType?: string
providerName?: string
baseUrl?: string
modelId?: string
apiKey?: string
}
export interface BootstrapConfigInput {
gatewayPort: number
gatewayToken: string
browserosServerPort?: number
providerType?: string
providerName?: string
baseUrl?: string
modelId?: string
}
export interface EnvFileInput {
image?: string
port?: number
token: string
configDir: string
timezone?: string
providerKeys?: Record<string, string>
}
export interface ResolvedProviderConfig {
model?: string
providerKeys: Record<string, string>
models?: {
mode: 'merge'
providers: Record<string, Record<string, unknown>>
}
}
function hasBuiltinProvider(providerType?: string): providerType is string {
return !!providerType && providerType in PROVIDER_ENV_MAP
}
/**
* OpenRouter's public slugs use dots for version numbers
* (e.g. `anthropic/claude-haiku-4.5`), but openclaw's model registry expects
* dashes (`claude-haiku-4-5`). Passing the dotted form makes openclaw fail
* the registry lookup silently and the agent turn completes with zero
* payloads. Rewrite dots to dashes for openrouter model ids only.
*/
function normalizeBuiltinModelId(
providerType: string,
modelId: string,
): string {
if (providerType !== 'openrouter') return modelId
return modelId.replace(/\./g, '-')
}
export function deriveOpenClawProviderId(providerInput: {
providerType?: string
providerName?: string
baseUrl?: string
}): string {
const source =
providerInput.providerName?.trim() ||
providerInput.baseUrl?.trim() ||
providerInput.providerType?.trim() ||
'custom-provider'
const candidate = source
.toLowerCase()
.replace(/^https?:\/\//, '')
.replace(/[^a-z0-9]+/g, '-')
.replace(/^-|-$/g, '')
return candidate || 'custom-provider'
}
export function deriveOpenClawApiKeyEnvVar(providerId: string): string {
return `${providerId.toUpperCase().replace(/-/g, '_')}_API_KEY`
}
export function resolveProviderConfig(
input: OpenClawProviderInput,
): ResolvedProviderConfig {
if (!input.providerType) {
return { providerKeys: {} }
}
if (hasBuiltinProvider(input.providerType)) {
const providerKeys: Record<string, string> = {}
if (input.apiKey) {
providerKeys[PROVIDER_ENV_MAP[input.providerType]] = input.apiKey
}
const normalizedModelId = input.modelId
? normalizeBuiltinModelId(input.providerType, input.modelId)
: undefined
return {
providerKeys,
model: normalizedModelId
? `${input.providerType}/${normalizedModelId}`
: undefined,
}
}
if (!input.baseUrl) {
return { providerKeys: {} }
}
const providerId = deriveOpenClawProviderId(input)
const apiKeyEnvVar = deriveOpenClawApiKeyEnvVar(providerId)
const providerKeys: Record<string, string> = {}
if (input.apiKey) {
providerKeys[apiKeyEnvVar] = input.apiKey
}
const providerConfig: Record<string, unknown> = {
baseUrl: input.baseUrl,
apiKey: `\${${apiKeyEnvVar}}`,
api: 'openai-completions',
}
if (input.modelId) {
providerConfig.models = [{ id: input.modelId, name: input.modelId }]
}
return {
providerKeys,
model: input.modelId ? `${providerId}/${input.modelId}` : undefined,
models: {
mode: 'merge',
providers: {
[providerId]: providerConfig,
},
},
}
}
export function buildBootstrapConfig(
input: BootstrapConfigInput,
): Record<string, unknown> {
const serverPort = input.browserosServerPort ?? DEFAULT_PORTS.server
const provider = resolveProviderConfig(input)
const defaults: Record<string, unknown> = {
workspace: `${OPENCLAW_CONTAINER_HOME}/workspace`,
timeoutSeconds: 4200,
thinkingDefault: 'adaptive',
}
if (provider.model) {
defaults.model = { primary: provider.model }
}
const config: Record<string, unknown> = {
gateway: {
mode: 'local',
port: input.gatewayPort,
bind: 'lan',
auth: { mode: 'token', token: input.gatewayToken },
reload: { mode: 'restart' },
controlUi: {
allowInsecureAuth: true,
allowedOrigins: [
`http://127.0.0.1:${input.gatewayPort}`,
`http://localhost:${input.gatewayPort}`,
],
},
http: {
endpoints: {
chatCompletions: { enabled: true },
},
},
},
agents: { defaults },
tools: {
profile: 'full',
web: {
search: { provider: 'duckduckgo', enabled: true },
},
exec: {
host: 'gateway',
security: 'full',
ask: 'off',
},
},
cron: { enabled: true },
hooks: {
internal: {
enabled: true,
entries: {
'boot-md': { enabled: true },
'bootstrap-extra-files': { enabled: true },
'session-memory': { enabled: true },
},
},
},
mcp: {
servers: {
browseros: {
url: `http://host.containers.internal:${serverPort}/mcp`,
transport: 'streamable-http',
},
},
},
approvals: {
exec: { enabled: false },
},
skills: {
install: { nodeManager: 'bun' },
},
}
if (provider.models) {
config.models = provider.models
}
if (process.env.NODE_ENV === 'development') {
config.logging = { level: 'debug', consoleLevel: 'debug' }
}
return config
}
export function buildEnvFile(input: EnvFileInput): string {
const lines: string[] = [
`OPENCLAW_IMAGE=${input.image ?? OPENCLAW_IMAGE}`,
`OPENCLAW_GATEWAY_PORT=${input.port ?? OPENCLAW_GATEWAY_PORT}`,
`OPENCLAW_GATEWAY_TOKEN=${input.token}`,
`OPENCLAW_CONFIG_DIR=${input.configDir}`,
`TZ=${input.timezone ?? Intl.DateTimeFormat().resolvedOptions().timeZone}`,
]
if (input.providerKeys) {
for (const [key, value] of Object.entries(input.providerKeys)) {
lines.push(`${key}=${value}`)
}
}
return `${lines.join('\n')}\n`
}
export function resolveProviderKeys(
input: OpenClawProviderInput,
): Record<string, string> {
return resolveProviderConfig(input).providerKeys
}
export function resolveProviderModel(
input: OpenClawProviderInput,
): string | undefined {
return resolveProviderConfig(input).model
}

View File

@@ -0,0 +1,73 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { join } from 'node:path'
const STATE_DIR_NAME = '.openclaw'
export function getOpenClawStateDir(openclawDir: string): string {
return join(openclawDir, STATE_DIR_NAME)
}
export function getOpenClawStateConfigPath(openclawDir: string): string {
return join(getOpenClawStateDir(openclawDir), 'openclaw.json')
}
export function getOpenClawStateEnvPath(openclawDir: string): string {
return join(getOpenClawStateDir(openclawDir), '.env')
}
export function getHostWorkspaceDir(
openclawDir: string,
agentName: string,
): string {
return join(
getOpenClawStateDir(openclawDir),
agentName === 'main' ? 'workspace' : `workspace-${agentName}`,
)
}
export function mergeEnvContent(
current: string,
updates: Record<string, string>,
): { changed: boolean; content: string } {
if (Object.keys(updates).length === 0) {
return {
changed: false,
content: normalizeEnvContent(current),
}
}
const lines = current === '' ? [] : current.replace(/\r\n/g, '\n').split('\n')
const nextLines = [...lines]
let changed = false
for (const [key, value] of Object.entries(updates)) {
const replacement = `${key}=${value}`
const index = nextLines.findIndex((line) => line.startsWith(`${key}=`))
if (index === -1) {
nextLines.push(replacement)
changed = true
continue
}
if (nextLines[index] === replacement) {
continue
}
nextLines[index] = replacement
changed = true
}
const content = normalizeEnvContent(nextLines.join('\n'))
return {
changed: changed || content !== normalizeEnvContent(current),
content,
}
}
function normalizeEnvContent(content: string): string {
const trimmed = content.trim()
return trimmed ? `${trimmed}\n` : ''
}

View File

@@ -0,0 +1,529 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { createParser, type EventSourceMessage } from 'eventsource-parser'
import { OpenClawSessionNotFoundError } from './errors'
import type { OpenClawStreamEvent } from './openclaw-types'
export interface OpenClawChatHistoryMessage {
role: 'user' | 'assistant'
content: string
}
/**
* OpenAI-compatible content parts for multimodal user messages. OpenClaw's
* gateway accepts the standard `content: [{ type: 'text', ... }, { type:
* 'image_url', image_url: { url } }]` shape on /v1/chat/completions and
* routes it to whichever upstream provider the agent's model points at.
*/
export type OpenClawChatContentPart =
| { type: 'text'; text: string }
| {
type: 'image_url'
image_url: { url: string; detail?: 'auto' | 'low' | 'high' }
}
export interface OpenClawChatRequest {
agentId: string
sessionKey: string
message: string
// When present, sent as the user message's `content` array verbatim. The
// legacy string `message` is folded into a leading text part if no text
// part is present in `messageParts`.
messageParts?: OpenClawChatContentPart[]
history?: OpenClawChatHistoryMessage[]
signal?: AbortSignal
}
export interface OpenClawSessionHistoryMessage {
role: 'user' | 'assistant' | 'system' | 'tool'
content: string
messageId?: string
messageSeq?: number
timestamp?: number
}
export interface OpenClawSessionHistory {
sessionKey: string
messages: OpenClawSessionHistoryMessage[]
cursor?: string | null
hasMore?: boolean
truncated?: boolean
}
export interface OpenClawSessionHistoryInput {
limit?: number
cursor?: string
signal?: AbortSignal
}
export type OpenClawSessionHistoryEvent =
| { type: 'history'; data: OpenClawSessionHistory }
| {
type: 'message'
data: {
sessionKey: string
message: OpenClawSessionHistoryMessage
messageId?: string
messageSeq: number
}
}
| { type: 'error'; data: { message: string } }
export class OpenClawHttpClient {
constructor(
private readonly hostPort: number,
private readonly getToken: () => Promise<string>,
) {}
async streamChat(
input: OpenClawChatRequest,
): Promise<ReadableStream<OpenClawStreamEvent>> {
const response = await this.fetchChat(input)
const body = response.body
if (!body) {
throw new Error('OpenClaw chat response had no body')
}
return createEventStream(body, input.signal)
}
async getSessionHistory(
sessionKey: string,
input: OpenClawSessionHistoryInput = {},
): Promise<OpenClawSessionHistory> {
const response = await this.fetchSessionHistory(sessionKey, input, {})
return (await response.json()) as OpenClawSessionHistory
}
async streamSessionHistory(
sessionKey: string,
input: OpenClawSessionHistoryInput = {},
): Promise<ReadableStream<OpenClawSessionHistoryEvent>> {
const response = await this.fetchSessionHistory(sessionKey, input, {
Accept: 'text/event-stream',
})
const body = response.body
if (!body) {
throw new Error('OpenClaw session history stream had no body')
}
return createHistoryEventStream(body, input.signal)
}
async isAuthenticated(): Promise<boolean> {
try {
const token = await this.getToken()
const response = await fetch(
`http://127.0.0.1:${this.hostPort}/v1/models`,
{
method: 'GET',
headers: {
Authorization: `Bearer ${token}`,
},
},
)
return response.ok
} catch {
return false
}
}
private async fetchChat(input: OpenClawChatRequest): Promise<Response> {
const token = await this.getToken()
const userContent = buildUserContent(input)
const response = await fetch(
`http://127.0.0.1:${this.hostPort}/v1/chat/completions`,
{
method: 'POST',
headers: {
Authorization: `Bearer ${token}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
model: resolveAgentModel(input.agentId),
stream: true,
messages: [
...(input.history ?? []),
{ role: 'user', content: userContent },
],
user: `browseros:${input.agentId}:${input.sessionKey}`,
}),
signal: input.signal,
},
)
if (response.ok) {
return response
}
const detail = await response.text()
throw new Error(
detail || `OpenClaw chat failed with status ${response.status}`,
)
}
private async fetchSessionHistory(
sessionKey: string,
input: OpenClawSessionHistoryInput,
extraHeaders: Record<string, string>,
): Promise<Response> {
const token = await this.getToken()
const response = await fetch(
`http://127.0.0.1:${this.hostPort}${buildHistoryPath(sessionKey, input)}`,
{
method: 'GET',
headers: {
Authorization: `Bearer ${token}`,
...extraHeaders,
},
signal: input.signal,
},
)
if (response.status === 404) {
throw new OpenClawSessionNotFoundError(sessionKey)
}
if (!response.ok) {
const detail = await response.text()
throw new Error(
detail ||
`OpenClaw session history failed with status ${response.status}`,
)
}
return response
}
}
function buildHistoryPath(
sessionKey: string,
input: OpenClawSessionHistoryInput,
): string {
const qs = new URLSearchParams()
if (input.limit !== undefined) qs.set('limit', String(input.limit))
if (input.cursor !== undefined) qs.set('cursor', input.cursor)
const suffix = qs.toString()
return `/sessions/${encodeURIComponent(sessionKey)}/history${
suffix ? `?${suffix}` : ''
}`
}
function resolveAgentModel(agentId: string): string {
return agentId === 'main' ? 'openclaw' : `openclaw/${agentId}`
}
/**
* Build the OpenAI-compatible `content` payload for the trailing user
* message. When the caller supplies multimodal parts via `messageParts`,
* use them as-is, ensuring at least one text part is present (we fold the
* legacy `message` string in as a leading text part if not). Otherwise,
* fall back to a plain string `content` so simple text-only sends keep
* the same wire shape we've always sent.
*/
function buildUserContent(
input: OpenClawChatRequest,
): string | OpenClawChatContentPart[] {
if (!input.messageParts || input.messageParts.length === 0) {
return input.message
}
const hasText = input.messageParts.some((p) => p.type === 'text')
if (hasText) return input.messageParts
const trimmed = input.message.trim()
if (!trimmed) return input.messageParts
return [{ type: 'text', text: input.message }, ...input.messageParts]
}
function createEventStream(
body: ReadableStream<Uint8Array>,
signal?: AbortSignal,
): ReadableStream<OpenClawStreamEvent> {
return new ReadableStream<OpenClawStreamEvent>({
start(controller) {
void pumpChatEvents(body, controller, signal)
},
})
}
async function pumpChatEvents(
body: ReadableStream<Uint8Array>,
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
signal?: AbortSignal,
): Promise<void> {
const reader = body.getReader()
const decoder = new TextDecoder()
let text = ''
let done = false
const parser = createParser({
onEvent(message) {
if (done) return
const nextText = updateAccumulatedText(message, text)
done = handleMessage(message, controller, nextText, done)
if (!done) {
text = nextText
}
},
})
try {
while (true) {
if (signal?.aborted) {
await reader.cancel()
done = true
controller.close()
return
}
const { done: streamDone, value } = await reader.read()
if (streamDone) break
parser.feed(decoder.decode(value, { stream: true }))
}
} catch (error) {
if (!done) {
controller.enqueue({
type: 'error',
data: {
message: error instanceof Error ? error.message : String(error),
},
})
done = true
controller.close()
}
} finally {
if (!done) {
controller.close()
}
reader.releaseLock()
}
}
function handleMessage(
message: EventSourceMessage,
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
text: string,
done: boolean,
): boolean {
if (message.data === '[DONE]') {
return finishStream(controller, text, done)
}
const chunk = parseChunk(message.data)
if (!chunk) {
controller.enqueue({
type: 'error',
data: { message: 'Failed to parse OpenClaw chat stream chunk' },
})
controller.close()
return true
}
for (const event of mapChunkToEvents(chunk)) {
controller.enqueue(event)
}
return hasFinishReason(chunk) ? finishStream(controller, text, done) : false
}
function updateAccumulatedText(
message: EventSourceMessage,
text: string,
): string {
const chunk = parseChunk(message.data)
if (!chunk) return text
let next = text
for (const choice of readChoices(chunk)) {
const delta = readDeltaText(choice)
if (delta) {
next += delta
}
}
return next
}
function finishStream(
controller: ReadableStreamDefaultController<OpenClawStreamEvent>,
text: string,
done: boolean,
): boolean {
if (!done) {
if (!text.trim()) {
controller.enqueue({
type: 'error',
data: {
message: "Agent couldn't generate a response. Please try again.",
},
})
controller.close()
return true
}
controller.enqueue({
type: 'done',
data: { text },
})
controller.close()
}
return true
}
function mapChunkToEvents(
chunk: Record<string, unknown>,
): OpenClawStreamEvent[] {
const events: OpenClawStreamEvent[] = []
for (const choice of readChoices(chunk)) {
const delta = readDeltaText(choice)
if (delta) {
events.push({
type: 'text-delta',
data: { text: delta },
})
}
}
return events
}
function hasFinishReason(chunk: Record<string, unknown>): boolean {
return readChoices(chunk).some((choice) => !!readFinishReason(choice))
}
function readChoices(
chunk: Record<string, unknown>,
): Array<Record<string, unknown>> {
const choices = chunk.choices
return Array.isArray(choices)
? choices.filter(
(choice): choice is Record<string, unknown> =>
!!choice && typeof choice === 'object',
)
: []
}
function readDeltaText(choice: Record<string, unknown>): string {
const delta = choice.delta
if (!delta || typeof delta !== 'object') return ''
const content = (delta as Record<string, unknown>).content
return typeof content === 'string' ? content : ''
}
function readFinishReason(choice: Record<string, unknown>): string | null {
const reason = choice.finish_reason
return typeof reason === 'string' && reason ? reason : null
}
function parseChunk(data: string): Record<string, unknown> | null {
try {
return JSON.parse(data) as Record<string, unknown>
} catch {
return null
}
}
function createHistoryEventStream(
body: ReadableStream<Uint8Array>,
signal?: AbortSignal,
): ReadableStream<OpenClawSessionHistoryEvent> {
return new ReadableStream<OpenClawSessionHistoryEvent>({
start(controller) {
void pumpHistoryEvents(body, controller, signal)
},
})
}
async function pumpHistoryEvents(
body: ReadableStream<Uint8Array>,
controller: ReadableStreamDefaultController<OpenClawSessionHistoryEvent>,
signal?: AbortSignal,
): Promise<void> {
const reader = body.getReader()
const decoder = new TextDecoder()
let closed = false
const close = () => {
if (closed) return
closed = true
controller.close()
}
const parser = createParser({
onEvent(message) {
if (closed) return
const event = toHistoryEvent(message)
if (!event) return
controller.enqueue(event)
if (event.type === 'error') close()
},
})
const onAbort = () => {
void reader.cancel().catch(() => {})
close()
}
signal?.addEventListener('abort', onAbort, { once: true })
try {
while (true) {
if (signal?.aborted) {
await reader.cancel()
close()
return
}
const { done, value } = await reader.read()
if (done) break
parser.feed(decoder.decode(value, { stream: true }))
}
} catch (error) {
if (!closed) {
controller.enqueue({
type: 'error',
data: {
message: error instanceof Error ? error.message : String(error),
},
})
close()
}
} finally {
signal?.removeEventListener('abort', onAbort)
close()
reader.releaseLock()
}
}
function toHistoryEvent(
message: EventSourceMessage,
): OpenClawSessionHistoryEvent | null {
if (!message.event) return null
const payload = parseChunk(message.data)
if (!payload) return null
if (message.event === 'history') {
return {
type: 'history',
data: payload as unknown as OpenClawSessionHistory,
}
}
if (message.event === 'message') {
return {
type: 'message',
data: payload as unknown as {
sessionKey: string
message: OpenClawSessionHistoryMessage
messageId?: string
messageSeq: number
},
}
}
if (message.event === 'error') {
const errMessage =
typeof payload.message === 'string'
? payload.message
: 'OpenClaw session history stream error'
return { type: 'error', data: { message: errMessage } }
}
return null
}

View File

@@ -0,0 +1,667 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { existsSync, readdirSync, readFileSync, statSync } from 'node:fs'
import { resolve } from 'node:path'
// ---------------------------------------------------------------------------
// Types for raw JSONL line parsing (matches OpenClaw's internal format)
// ---------------------------------------------------------------------------
interface PiContentBlock {
type: string
text?: string
// OpenClaw stores reasoning blocks as { type: 'thinking', thinking: '...' }
// — the prose lives on a `thinking` field, not `text`.
thinking?: string
id?: string
name?: string
arguments?: Record<string, unknown>
// OpenAI-shaped image blocks: { type: 'image_url', image_url: { url } }.
// The data: URL carries mediaType + base64 in one string.
image_url?: { url?: string; detail?: string }
// Anthropic-shaped image blocks: { type: 'image', source: { type:
// 'base64', media_type, data } } and the simpler { type: 'image', data }
// variant the gateway emits on tool results.
source?: { type?: string; media_type?: string; data?: string }
data?: string
media_type?: string
mediaType?: string
}
interface PiMessage {
role?: 'user' | 'assistant' | 'toolResult'
content?: PiContentBlock[]
stopReason?: string
errorMessage?: string
usage?: {
input?: number
output?: number
cost?: {
total?: number
}
}
model?: string
provider?: string
toolCallId?: string
toolName?: string
isError?: boolean
}
interface PiLine {
type: string
id?: string
timestamp?: string
message?: PiMessage
provider?: string
modelId?: string
thinkingLevel?: string
summary?: string
firstKeptEntryId?: string
tokensBefore?: number
}
interface SessionsJsonEntry {
sessionId?: string
updatedAt?: number
[k: string]: unknown
}
type SessionsJson = Record<string, SessionsJsonEntry>
// ---------------------------------------------------------------------------
// Public types
// ---------------------------------------------------------------------------
export type ClawEventType =
| 'user.message'
| 'user.attachment'
| 'agent.message'
| 'agent.thinking'
| 'agent.tool_use'
| 'agent.tool_result'
| 'session.model_change'
| 'session.thinking_level_change'
| 'session.compaction'
export interface ClawAttachmentInfo {
kind: 'image' | 'file'
mediaType: string
// For images we always emit a data: URL so downstream consumers don't
// have to reconstruct it. `name` is best-effort (JSONL rarely carries
// a filename for inline image content blocks).
dataUrl?: string
name?: string
}
export interface ClawEvent {
eventId: string
type: ClawEventType
content: string
createdAt: number
tokensIn?: number
tokensOut?: number
costUsd?: number
model?: string
toolName?: string
toolCallId?: string
toolArguments?: Record<string, unknown>
isError?: boolean
attachment?: ClawAttachmentInfo
}
export interface JsonlSessionEntry {
key: string
sessionId: string
updatedAt: number
}
export interface JsonlSessionStats {
userTurns: number
assistantMessages: number
toolCalls: number
totalCostUsd: number
totalTokensIn: number
totalTokensOut: number
}
// ---------------------------------------------------------------------------
// Reader
// ---------------------------------------------------------------------------
/**
* Reads OpenClaw's per-session JSONL files directly from the host filesystem.
* OpenClaw is the sole writer — this reader never modifies the files.
*
* Path layout on the host (via Lima virtiofs mount):
* <stateRoot>/agents/<agentId>/sessions/sessions.json
* <stateRoot>/agents/<agentId>/sessions/<piSessionId>.jsonl
*/
export class OpenClawJsonlReader {
constructor(private readonly stateRoot: string) {}
/** List all sessions for an agent by reading sessions.json. */
listSessions(agentId: string): JsonlSessionEntry[] {
const sessionsJson = this.readSessionsJson(agentId)
if (!sessionsJson) return []
const entries: JsonlSessionEntry[] = []
for (const [key, entry] of Object.entries(sessionsJson)) {
if (typeof entry.sessionId === 'string') {
entries.push({
key,
sessionId: entry.sessionId,
updatedAt: typeof entry.updatedAt === 'number' ? entry.updatedAt : 0,
})
}
}
return entries.sort((a, b) => b.updatedAt - a.updatedAt)
}
/** List all agent IDs by scanning the agents directory. */
listAgents(): string[] {
try {
const entries = readdirSync(this.safePath('agents'), {
withFileTypes: true,
})
return entries.filter((e) => e.isDirectory()).map((e) => e.name)
} catch {
return []
}
}
/**
* Read and parse all events from a session's JSONL file.
*
* Uses resolveJsonlPath() which handles a known OpenClaw quirk: the
* Pi session ID recorded in sessions.json can drift from the actual
* JSONL filename after context compaction or session restart. When the
* mapped ID doesn't match a file on disk, we fall back to the most
* recently modified JSONL in the agent's sessions directory.
*/
listBySession(agentId: string, sessionKey: string): ClawEvent[] {
const filePath = this.resolveJsonlPath(agentId, sessionKey)
if (!filePath) return []
let raw: string
try {
raw = readFileSync(filePath, 'utf8')
} catch {
return []
}
const events: ClawEvent[] = []
for (const line of raw.split('\n')) {
if (!line.trim()) continue
let parsed: PiLine
try {
parsed = JSON.parse(line) as PiLine
} catch {
// Skip malformed lines — a partial line at the tail is possible
// if OpenClaw is mid-write.
continue
}
for (const event of mapLineToEvents(parsed)) {
events.push(event)
}
}
return events
}
/** Get the latest assistant message from a session. */
latestAgentMessage(
agentId: string,
sessionKey: string,
): ClawEvent | undefined {
const events = this.listBySession(agentId, sessionKey)
for (let i = events.length - 1; i >= 0; i--) {
if (events[i]?.type === 'agent.message') return events[i]
}
return undefined
}
/** Count user turns in a session. */
countUserTurns(agentId: string, sessionKey: string): number {
const events = this.listBySession(agentId, sessionKey)
let n = 0
for (const e of events) {
if (e.type === 'user.message') n++
}
return n
}
/** Aggregate stats for a session. */
getSessionStats(agentId: string, sessionKey: string): JsonlSessionStats {
const events = this.listBySession(agentId, sessionKey)
const stats: JsonlSessionStats = {
userTurns: 0,
assistantMessages: 0,
toolCalls: 0,
totalCostUsd: 0,
totalTokensIn: 0,
totalTokensOut: 0,
}
for (const e of events) {
if (e.type === 'user.message') stats.userTurns++
if (e.type === 'agent.message') {
stats.assistantMessages++
if (e.costUsd) stats.totalCostUsd += e.costUsd
if (e.tokensIn) stats.totalTokensIn += e.tokensIn
if (e.tokensOut) stats.totalTokensOut += e.tokensOut
}
if (e.type === 'agent.tool_use') stats.toolCalls++
}
return stats
}
// ── Private helpers ─────────────────────────────────────────────────
/**
* Ensure a resolved path stays within stateRoot to prevent path traversal
* via crafted agentId or sessionId values containing ".." segments.
*/
private safePath(...segments: string[]): string {
const resolved = resolve(this.stateRoot, ...segments)
const root = resolve(this.stateRoot)
if (!resolved.startsWith(`${root}/`) && resolved !== root) {
throw new Error(`Path traversal blocked: ${segments.join('/')}`)
}
return resolved
}
private readSessionsJson(agentId: string): SessionsJson | null {
const filePath = this.safePath(
'agents',
agentId,
'sessions',
'sessions.json',
)
try {
const raw = readFileSync(filePath, 'utf8')
return JSON.parse(raw) as SessionsJson
} catch {
return null
}
}
/**
* Resolve the path to a session's JSONL file. Tries the sessions.json
* mapping first (fast), then falls back to scanning the directory for
* the most recently modified JSONL file when the mapped ID doesn't
* match an actual file on disk.
*
* This fallback handles a known OpenClaw behavior where the Pi session
* ID in sessions.json can become stale after context compaction or
* session restart — the JSONL file on disk has a different UUID than
* what sessions.json records.
*/
private resolveJsonlPath(agentId: string, sessionKey: string): string | null {
const sessionsJson = this.readSessionsJson(agentId)
if (!sessionsJson) return null
// Try exact key match in sessions.json
let resolvedId: string | undefined
const entry = sessionsJson[sessionKey]
if (entry && typeof entry.sessionId === 'string') {
resolvedId = entry.sessionId
}
// Try matching by scanning all keys (handles key format variations)
if (!resolvedId) {
for (const [key, value] of Object.entries(sessionsJson)) {
if (key === sessionKey || key.endsWith(`:${sessionKey}`)) {
if (typeof value.sessionId === 'string') {
resolvedId = value.sessionId
break
}
}
}
}
// If we found a sessionId and the file exists, use it
if (resolvedId) {
const path = this.safePath(
'agents',
agentId,
'sessions',
`${resolvedId}.jsonl`,
)
if (existsSync(path)) return path
}
// Fallback: scan the sessions directory for the most recent JSONL
// file. This handles stale sessions.json entries where the Pi
// session ID doesn't match the actual file on disk.
return this.findMostRecentJsonl(agentId)
}
/**
* Scan the sessions directory and return the path to the most recently
* modified JSONL file. Used as a fallback when sessions.json points to
* a non-existent file.
*/
private findMostRecentJsonl(agentId: string): string | null {
let sessionsDir: string
try {
sessionsDir = this.safePath('agents', agentId, 'sessions')
} catch {
return null
}
let names: string[]
try {
names = readdirSync(sessionsDir).filter(
(n): n is string => typeof n === 'string' && n.endsWith('.jsonl'),
)
} catch {
return null
}
let best: { path: string; mtime: number } | null = null
for (const name of names) {
const fullPath = this.safePath('agents', agentId, 'sessions', name)
try {
const st = statSync(fullPath)
if (!best || st.mtimeMs > best.mtime) {
best = { path: fullPath, mtime: st.mtimeMs }
}
} catch {}
}
return best?.path ?? null
}
}
// ---------------------------------------------------------------------------
// JSONL line → ClawEvent mapping
// ---------------------------------------------------------------------------
function mapLineToEvents(line: PiLine): ClawEvent[] {
const eventId = line.id ?? ''
const createdAt = line.timestamp ? Date.parse(line.timestamp) : Date.now()
if (line.type === 'model_change') {
const model = combineModel(line.provider, line.modelId)
if (!model) return []
return [
{
eventId,
type: 'session.model_change',
content: model,
createdAt,
model,
},
]
}
if (line.type === 'thinking_level_change') {
return [
{
eventId,
type: 'session.thinking_level_change',
content: line.thinkingLevel ?? 'unknown',
createdAt,
},
]
}
if (line.type === 'compaction') {
return [
{
eventId,
type: 'session.compaction',
content: line.summary ?? '(compacted)',
createdAt,
},
]
}
if (line.type !== 'message' || !line.message) return []
return mapMessageToEvents(line.message, eventId, createdAt)
}
function mapMessageToEvents(
msg: PiMessage,
eventId: string,
createdAt: number,
): ClawEvent[] {
if (msg.role === 'user') {
return mapUserMessage(msg, eventId, createdAt)
}
if (msg.role === 'assistant') {
return mapAssistantMessage(msg, eventId, createdAt)
}
if (msg.role === 'toolResult') {
const text = extractText(msg.content)
return [
{
eventId,
type: 'agent.tool_result',
content: text || '(no output)',
createdAt,
toolName: msg.toolName,
toolCallId: msg.toolCallId,
isError: msg.isError,
},
]
}
return []
}
/**
* Build events for a user JSONL message. Each image content block becomes
* a separate `user.attachment` event ordered before the `user.message`
* text event so downstream accumulators (in jsonlEventsToHistoryItems)
* can flush attachments onto the message they arrived with.
*/
function mapUserMessage(
msg: PiMessage,
eventId: string,
createdAt: number,
): ClawEvent[] {
const events: ClawEvent[] = []
const text = extractText(msg.content)
if (msg.content) {
let attachmentIdx = 0
for (const block of msg.content) {
const attachment = extractImageAttachment(block)
if (!attachment) continue
events.push({
eventId: `${eventId}:attachment:${attachmentIdx}`,
type: 'user.attachment',
content: attachment.dataUrl ?? '',
createdAt,
attachment,
})
attachmentIdx++
}
}
if (text) {
events.push({ eventId, type: 'user.message', content: text, createdAt })
} else if (events.length > 0) {
// User sent only attachments and no caption — synthesize an empty
// user.message so downstream pipelines that gate on user.message still
// see a turn boundary.
events.push({ eventId, type: 'user.message', content: '', createdAt })
}
return events
}
/**
* Extract a normalised image attachment from a single content block.
* Handles all three shapes the OpenClaw gateway round-trips:
* - OpenAI: `{ type: 'image_url', image_url: { url } }` (data: URL)
* - Anthropic: `{ type: 'image', source: { type: 'base64', media_type, data } }`
* - Bare: `{ type: 'image', data: '<base64>' }` (used by tool-result outputs)
*/
function extractImageAttachment(
block: PiContentBlock,
): ClawAttachmentInfo | null {
if (block.type === 'image_url') {
const url = block.image_url?.url
if (typeof url !== 'string' || !url.startsWith('data:')) return null
const mediaType =
url.slice(5, url.indexOf(';')).trim() || 'application/octet-stream'
return { kind: 'image', mediaType, dataUrl: url }
}
if (block.type === 'image') {
const sourceData = block.source?.data
const sourceMediaType =
block.source?.media_type ?? block.media_type ?? block.mediaType
const bareData = block.data
if (typeof sourceData === 'string' && typeof sourceMediaType === 'string') {
return {
kind: 'image',
mediaType: sourceMediaType,
dataUrl: `data:${sourceMediaType};base64,${sourceData}`,
}
}
if (typeof bareData === 'string') {
const mediaType =
typeof sourceMediaType === 'string' ? sourceMediaType : 'image/png'
return {
kind: 'image',
mediaType,
dataUrl: `data:${mediaType};base64,${bareData}`,
}
}
}
return null
}
function mapAssistantMessage(
msg: PiMessage,
eventId: string,
createdAt: number,
): ClawEvent[] {
const events: ClawEvent[] = []
const text = extractText(msg.content)
if (msg.content) {
let thinkingIdx = 0
let toolIdx = 0
for (const block of msg.content) {
if (block.type === 'thinking') {
const thinkingText =
(typeof block.thinking === 'string' && block.thinking) ||
(typeof block.text === 'string' && block.text) ||
''
if (thinkingText.length > 0) {
events.push({
eventId: `${eventId}:thinking:${thinkingIdx}`,
type: 'agent.thinking',
content: thinkingText,
createdAt,
})
thinkingIdx++
}
}
if (block.type === 'toolCall' && block.name) {
events.push({
eventId: `${eventId}:tool:${block.id ?? toolIdx}`,
type: 'agent.tool_use',
content: block.name,
createdAt,
toolName: block.name,
toolCallId: block.id,
toolArguments: block.arguments,
})
toolIdx++
}
}
}
if (text) {
events.push({
eventId,
type: 'agent.message',
content: text,
createdAt,
tokensIn: msg.usage?.input,
tokensOut: msg.usage?.output,
costUsd: msg.usage?.cost?.total,
model: combineModel(msg.provider, msg.model),
})
}
return events
}
function extractText(blocks: PiContentBlock[] | undefined): string {
if (!blocks || blocks.length === 0) return ''
const parts: string[] = []
for (const block of blocks) {
if (block.type === 'text' && typeof block.text === 'string') {
parts.push(block.text)
}
}
return parts.join('')
}
function combineModel(
provider: string | undefined,
model: string | undefined,
): string | undefined {
if (!model) return undefined
return provider ? `${provider}/${model}` : model
}
// ---------------------------------------------------------------------------
// Tool activity summary
// ---------------------------------------------------------------------------
const TOOL_DESCRIPTIONS: Record<string, (count: number) => string> = {
browser_navigate: (n) => `Browsed ${n} page${n !== 1 ? 's' : ''}`,
browser_take_screenshot: (n) => `Took ${n} screenshot${n !== 1 ? 's' : ''}`,
browser_click: (n) => `Clicked ${n} element${n !== 1 ? 's' : ''}`,
browser_fill: (n) => `Filled ${n} field${n !== 1 ? 's' : ''}`,
browser_type: (n) => `Typed in ${n} field${n !== 1 ? 's' : ''}`,
google_calendar_list_events: (n) =>
n > 1 ? `Checked calendar ${n} times` : 'Checked calendar',
gmail_search: (n) => (n > 1 ? `Searched email ${n} times` : 'Searched email'),
gmail_send: (n) => `Sent ${n} email${n !== 1 ? 's' : ''}`,
slack_post_message: (n) => `Sent ${n} Slack message${n !== 1 ? 's' : ''}`,
file_write: (n) => `Wrote ${n} file${n !== 1 ? 's' : ''}`,
file_read: (n) => `Read ${n} file${n !== 1 ? 's' : ''}`,
}
function defaultToolDescription(toolName: string, count: number): string {
const short = toolName
.replace(/^(browser_|google_|mcp_)/, '')
.replaceAll('_', ' ')
return count > 1 ? `Used ${short} ${count} times` : `Used ${short}`
}
/**
* Convert raw tool-use events into a human-readable activity summary.
*
* Example output: "Browsed 3 pages, took 2 screenshots"
*/
export function summarizeToolActivity(events: ClawEvent[]): string | null {
const toolCounts = new Map<string, number>()
for (const e of events) {
if (e.type === 'agent.tool_use' && e.toolName) {
toolCounts.set(e.toolName, (toolCounts.get(e.toolName) ?? 0) + 1)
}
}
if (toolCounts.size === 0) return null
const parts: string[] = []
for (const [tool, count] of toolCounts) {
const describe = TOOL_DESCRIPTIONS[tool]
parts.push(describe ? describe(count) : defaultToolDescription(tool, count))
}
return parts.join(', ')
}

View File

@@ -0,0 +1,276 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Connects to the OpenClaw gateway's WebSocket control plane and pipes
* chat broadcast events into a ClawSession state machine. The observer
* is a transport layer only — it handles the WS connection lifecycle
* (connect, handshake, reconnect) and delegates all state management
* to ClawSession.
*/
import WebSocket from 'ws'
import { logger } from '../../../lib/logger'
import type { ClawSession } from './claw-session'
// ---------------------------------------------------------------------------
// Protocol types (subset of OpenClaw gateway protocol v3)
// ---------------------------------------------------------------------------
const PROTOCOL_VERSION = 3
const HANDSHAKE_REQUEST_ID = 'connect'
const RECONNECT_DELAY_MS = 5_000
const CONNECT_TIMEOUT_MS = 10_000
interface RequestFrame {
type: 'req'
id: string
method: string
params: Record<string, unknown>
}
type IncomingFrame =
| { type: 'res'; id: string; ok: true; payload?: unknown }
| {
type: 'res'
id: string
ok: false
error: { code: string; message: string }
}
| { type: 'event'; event: string; payload?: unknown }
// ---------------------------------------------------------------------------
// Observer
// ---------------------------------------------------------------------------
export class OpenClawObserver {
private ws: WebSocket | null = null
private reconnectTimer: ReturnType<typeof setTimeout> | null = null
private connected = false
private closed = false
private gatewayUrl: string | null = null
private gatewayToken: string | null = null
constructor(private readonly session: ClawSession) {}
/** Start observing the gateway at the given URL with the given token. */
connect(gatewayUrl: string, token: string): void {
this.gatewayUrl = gatewayUrl
this.gatewayToken = token
this.closed = false
this.doConnect()
}
/** Stop observing and close the WebSocket. */
disconnect(): void {
this.closed = true
this.clearReconnect()
if (this.ws) {
try {
this.ws.close()
} catch {}
this.ws = null
}
this.connected = false
}
/** Whether the observer has an active WS connection. */
isConnected(): boolean {
return this.connected
}
// ── Private ─────────────────────────────────────────────────────────
private doConnect(): void {
if (this.closed || !this.gatewayUrl || !this.gatewayToken) return
const wsUrl = this.gatewayUrl
.replace(/^http:\/\//, 'ws://')
.replace(/^https:\/\//, 'wss://')
logger.debug('OpenClaw observer connecting', { url: wsUrl })
const ws = new WebSocket(wsUrl)
this.ws = ws
const connectTimeout = setTimeout(() => {
logger.warn('OpenClaw observer handshake timeout')
ws.terminate()
}, CONNECT_TIMEOUT_MS)
let handshakeSent = false
ws.on('message', (raw) => {
let frame: IncomingFrame
try {
frame = JSON.parse(raw.toString('utf8')) as IncomingFrame
} catch {
return
}
// The gateway sends a connect.challenge event before accepting
// the connect request. Send the handshake after receiving it.
if (
frame.type === 'event' &&
frame.event === 'connect.challenge' &&
!handshakeSent
) {
handshakeSent = true
const connectReq: RequestFrame = {
type: 'req',
id: HANDSHAKE_REQUEST_ID,
method: 'connect',
params: {
minProtocol: PROTOCOL_VERSION,
maxProtocol: PROTOCOL_VERSION,
client: {
id: 'openclaw-tui',
displayName: 'browseros-observer',
version: '1.0.0',
platform: 'node',
mode: 'ui',
},
role: 'operator',
scopes: ['operator.read'],
auth: { token: this.gatewayToken },
},
}
ws.send(JSON.stringify(connectReq))
return
}
// Handshake response
if (frame.type === 'res' && frame.id === HANDSHAKE_REQUEST_ID) {
clearTimeout(connectTimeout)
if (frame.ok) {
this.connected = true
logger.info('OpenClaw observer connected')
} else {
logger.warn('OpenClaw observer handshake failed', {
error: frame.error,
})
ws.close()
}
return
}
// Broadcast events (only process after handshake completes)
if (frame.type === 'event' && this.connected) {
this.handleEvent(frame.event, frame.payload)
}
})
ws.on('close', () => {
clearTimeout(connectTimeout)
this.connected = false
this.ws = null
// Reset any agents stuck in "working" to "unknown" — we missed
// the final/end event because the WS closed mid-task. The
// ClawSession will re-infer correct state from JSONL when the
// observer reconnects and ensureObserverConnected() re-seeds.
for (const [agentId, state] of this.session.getAllStates()) {
if (state.status === 'working') {
this.session.transition(agentId, 'unknown')
}
}
if (!this.closed) {
logger.debug('OpenClaw observer disconnected, scheduling reconnect')
this.scheduleReconnect()
}
})
ws.on('error', (err) => {
clearTimeout(connectTimeout)
logger.debug('OpenClaw observer WS error', {
message: err.message,
})
})
}
private handleEvent(eventName: string, payload: unknown): void {
if (eventName === 'chat') {
this.handleChatEvent(payload)
}
}
/**
* Parse a gateway chat broadcast event and transition the ClawSession
* state machine accordingly.
*/
private handleChatEvent(payload: unknown): void {
if (!payload || typeof payload !== 'object') return
const p = payload as Record<string, unknown>
const sessionKey = typeof p.sessionKey === 'string' ? p.sessionKey : null
const state = typeof p.state === 'string' ? p.state : null
if (!sessionKey || !state) return
const agentId = extractAgentId(sessionKey)
if (!agentId) return
if (state === 'delta' || state === 'streaming') {
this.session.transition(agentId, 'working', {
sessionKey,
currentTool: extractToolName(p),
})
} else if (state === 'final' || state === 'end') {
this.session.transition(agentId, 'idle', { sessionKey })
} else if (state === 'error') {
const errorMsg =
typeof p.errorMessage === 'string'
? p.errorMessage
: typeof p.error === 'string'
? p.error
: 'Unknown error'
this.session.transition(agentId, 'error', { sessionKey, error: errorMsg })
}
}
private scheduleReconnect(): void {
this.clearReconnect()
this.reconnectTimer = setTimeout(() => {
this.reconnectTimer = null
this.doConnect()
}, RECONNECT_DELAY_MS)
}
private clearReconnect(): void {
if (this.reconnectTimer) {
clearTimeout(this.reconnectTimer)
this.reconnectTimer = null
}
}
}
// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------
/**
* Extract agentId from an OpenClaw session key.
* Format: "agent:<agentId>:..." — we take the segment after "agent:".
*/
function extractAgentId(sessionKey: string): string | null {
if (!sessionKey.startsWith('agent:')) return null
const colonIdx = sessionKey.indexOf(':', 6)
if (colonIdx === -1) return sessionKey.slice(6)
return sessionKey.slice(6, colonIdx)
}
/**
* Try to extract a tool name from a chat event payload.
*/
function extractToolName(payload: Record<string, unknown>): string | null {
if (typeof payload.toolName === 'string') return payload.toolName
if (typeof payload.tool === 'string') return payload.tool
const content = payload.content
if (content && typeof content === 'object' && 'name' in content) {
const name = (content as Record<string, unknown>).name
if (typeof name === 'string') return name
}
return null
}

View File

@@ -0,0 +1,157 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
export const SUPPORTED_OPENCLAW_PROVIDERS = [
'openrouter',
'openai',
'anthropic',
'moonshot',
] as const
export type SupportedOpenClawProvider =
(typeof SUPPORTED_OPENCLAW_PROVIDERS)[number]
export interface CustomOpenClawProviderConfig {
providerId: string
apiKeyEnvVar: string
config: Record<string, unknown>
}
export interface ResolvedOpenClawProviderConfig {
envValues: Record<string, string>
model?: string
providerType?: SupportedOpenClawProvider
customProvider?: CustomOpenClawProviderConfig
}
const PROVIDER_ENV_VARS: Record<SupportedOpenClawProvider, string> = {
anthropic: 'ANTHROPIC_API_KEY',
moonshot: 'MOONSHOT_API_KEY',
openai: 'OPENAI_API_KEY',
openrouter: 'OPENROUTER_API_KEY',
}
export class UnsupportedOpenClawProviderError extends Error {
constructor(providerType: string) {
super(`Unsupported OpenClaw provider: ${providerType}`)
this.name = 'UnsupportedOpenClawProviderError'
}
}
export function isUnsupportedOpenClawProviderError(
error: unknown,
): error is UnsupportedOpenClawProviderError {
return (
error instanceof UnsupportedOpenClawProviderError ||
(error instanceof Error &&
error.name === 'UnsupportedOpenClawProviderError')
)
}
export function isSupportedOpenClawProvider(
providerType: string,
): providerType is SupportedOpenClawProvider {
return SUPPORTED_OPENCLAW_PROVIDERS.includes(
providerType as SupportedOpenClawProvider,
)
}
export function assertSupportedOpenClawProvider(
providerType?: string,
): SupportedOpenClawProvider | undefined {
if (!providerType) {
return undefined
}
if (!isSupportedOpenClawProvider(providerType)) {
throw new UnsupportedOpenClawProviderError(providerType)
}
return providerType
}
export function buildOpenClawModelRef(
providerType: SupportedOpenClawProvider,
modelId?: string,
): string | undefined {
return modelId ? `${providerType}/${modelId}` : undefined
}
export function deriveOpenClawProviderId(input: {
providerType?: string
providerName?: string
baseUrl?: string
}): string {
const source =
input.providerName?.trim() ||
input.baseUrl?.trim() ||
input.providerType?.trim() ||
'custom-provider'
const candidate = source
.toLowerCase()
.replace(/^https?:\/\//, '')
.replace(/[^a-z0-9]+/g, '-')
.replace(/^-|-$/g, '')
return candidate || 'custom-provider'
}
export function deriveOpenClawApiKeyEnvVar(providerId: string): string {
return `${providerId.toUpperCase().replace(/-/g, '_')}_API_KEY`
}
export function getOpenClawProviderEnvVar(
providerType: SupportedOpenClawProvider,
): string {
return PROVIDER_ENV_VARS[providerType]
}
export function resolveSupportedOpenClawProvider(input: {
providerType?: string
providerName?: string
baseUrl?: string
apiKey?: string
modelId?: string
}): ResolvedOpenClawProviderConfig {
if (!input.providerType) {
return { envValues: {} }
}
if (isSupportedOpenClawProvider(input.providerType)) {
const providerType = input.providerType
const envVar = getOpenClawProviderEnvVar(providerType)
return {
envValues: input.apiKey ? { [envVar]: input.apiKey } : {},
model: buildOpenClawModelRef(providerType, input.modelId),
providerType,
}
}
if (!input.baseUrl) {
throw new UnsupportedOpenClawProviderError(input.providerType)
}
const providerId = deriveOpenClawProviderId(input)
const apiKeyEnvVar = deriveOpenClawApiKeyEnvVar(providerId)
return {
envValues: input.apiKey ? { [apiKeyEnvVar]: input.apiKey } : {},
model: input.modelId ? `${providerId}/${input.modelId}` : undefined,
customProvider: {
providerId,
apiKeyEnvVar,
config: {
api: 'openai-completions',
baseUrl: input.baseUrl,
apiKey: `\${${apiKeyEnvVar}}`,
...(input.modelId
? {
models: [{ id: input.modelId, name: input.modelId }],
}
: {}),
},
},
}
}

View File

@@ -0,0 +1,18 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
export interface OpenClawStreamEvent {
type:
| 'text-delta'
| 'thinking'
| 'tool-start'
| 'tool-end'
| 'tool-output'
| 'lifecycle'
| 'done'
| 'error'
data: Record<string, unknown>
}

View File

@@ -1,285 +0,0 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Abstraction over the Podman CLI for container lifecycle management.
* Handles Podman machine init/start on macOS/Windows (where a Linux VM is required).
* On Linux, machine operations are no-ops since Podman runs natively.
*/
import { existsSync } from 'node:fs'
import { join } from 'node:path'
const isLinux = process.platform === 'linux'
const PODMAN_BUNDLE_PATH = ['bin', 'third_party', 'podman'] as const
export type LogFn = (msg: string) => void
function getPodmanBinaryName(platform: NodeJS.Platform): string {
return platform === 'win32' ? 'podman.exe' : 'podman'
}
export function resolveBundledPodmanPath(
resourcesDir?: string,
platform: NodeJS.Platform = process.platform,
): string | null {
if (!resourcesDir) return null
const bundledPath = join(
resourcesDir,
...PODMAN_BUNDLE_PATH,
getPodmanBinaryName(platform),
)
return existsSync(bundledPath) ? bundledPath : null
}
export class PodmanRuntime {
private podmanPath: string
private machineReady = false
constructor(config?: { podmanPath?: string }) {
this.podmanPath = config?.podmanPath ?? 'podman'
}
getPodmanPath(): string {
return this.podmanPath
}
async isPodmanAvailable(): Promise<boolean> {
try {
const proc = Bun.spawn([this.podmanPath, '--version'], {
stdout: 'ignore',
stderr: 'ignore',
})
return (await proc.exited) === 0
} catch {
return false
}
}
async getMachineStatus(): Promise<{
initialized: boolean
running: boolean
}> {
if (isLinux) return { initialized: true, running: true }
try {
const proc = Bun.spawn(
[this.podmanPath, 'machine', 'list', '--format', 'json'],
{ stdout: 'pipe', stderr: 'ignore' },
)
const output = await new Response(proc.stdout).text()
await proc.exited
const machines = JSON.parse(output) as Array<{
Running?: boolean
LastUp?: string
}>
if (!machines.length) return { initialized: false, running: false }
const machine = machines[0]
const running =
machine.Running === true || machine.LastUp === 'Currently running'
return { initialized: true, running }
} catch {
return { initialized: false, running: false }
}
}
async initMachine(onLog?: LogFn): Promise<void> {
if (isLinux) return
const proc = Bun.spawn(
[
this.podmanPath,
'machine',
'init',
'--cpus',
'2',
'--memory',
'2048',
'--disk-size',
'10',
],
{ stdout: 'ignore', stderr: 'pipe' },
)
await this.drainStderr(proc, onLog)
const code = await proc.exited
if (code !== 0)
throw new Error(`podman machine init failed with code ${code}`)
}
async startMachine(onLog?: LogFn): Promise<void> {
if (isLinux) return
const proc = Bun.spawn([this.podmanPath, 'machine', 'start'], {
stdout: 'ignore',
stderr: 'pipe',
})
await this.drainStderr(proc, onLog)
const code = await proc.exited
if (code !== 0)
throw new Error(`podman machine start failed with code ${code}`)
}
async stopMachine(): Promise<void> {
if (isLinux) return
const proc = Bun.spawn([this.podmanPath, 'machine', 'stop'], {
stdout: 'ignore',
stderr: 'ignore',
})
const code = await proc.exited
if (code !== 0)
throw new Error(`podman machine stop failed with code ${code}`)
this.machineReady = false
}
async ensureReady(onLog?: LogFn): Promise<void> {
if (this.machineReady) return
const status = await this.getMachineStatus()
if (!status.initialized) {
onLog?.('Initializing Podman machine...')
await this.initMachine(onLog)
}
if (!status.running) {
onLog?.('Starting Podman machine...')
await this.startMachine(onLog)
}
this.machineReady = true
}
async runCommand(
args: string[],
options?: {
cwd?: string
env?: Record<string, string>
onOutput?: (line: string) => void
},
): Promise<number> {
const useStreaming = !!options?.onOutput
const proc = Bun.spawn([this.podmanPath, ...args], {
cwd: options?.cwd,
env: options?.env ? { ...process.env, ...options.env } : undefined,
stdout: useStreaming ? 'pipe' : 'ignore',
stderr: useStreaming ? 'pipe' : 'ignore',
})
if (options?.onOutput) {
await Promise.all([
this.drainStream(proc.stdout ?? null, options.onOutput),
this.drainStream(proc.stderr ?? null, options.onOutput),
])
}
return proc.exited
}
/**
* Follow container logs. Returns a stop function that terminates the
* underlying `podman logs -f` process. Each output line is passed to
* onLine as-is.
*/
tailContainerLogs(containerName: string, onLine: LogFn): () => void {
const proc = Bun.spawn(
[this.podmanPath, 'logs', '-f', '--tail', '0', containerName],
{ stdout: 'pipe', stderr: 'pipe' },
)
void this.drainStream(proc.stdout ?? null, onLine)
void this.drainStream(proc.stderr ?? null, onLine)
let stopped = false
return () => {
if (stopped) return
stopped = true
try {
proc.kill()
} catch {
// process may already be gone
}
}
}
/**
* Lists running container names. Used to check whether non-BrowserOS
* containers are running before stopping the Podman machine.
*/
async listRunningContainers(): Promise<string[]> {
const proc = Bun.spawn([this.podmanPath, 'ps', '--format', '{{.Names}}'], {
stdout: 'pipe',
stderr: 'ignore',
})
const output = await new Response(proc.stdout).text()
await proc.exited
return output
.trim()
.split('\n')
.filter((name) => name.trim())
}
private async drainStderr(
proc: {
stderr: ReadableStream<Uint8Array> | null
exited: Promise<number>
},
onLog?: LogFn,
): Promise<void> {
if (!onLog || !proc.stderr) return
await this.drainStream(proc.stderr, onLog)
}
private async drainStream(
stream: ReadableStream<Uint8Array> | null,
onLine: (line: string) => void,
): Promise<void> {
if (!stream) return
const reader = stream.getReader()
const decoder = new TextDecoder()
let buffer = ''
while (true) {
const { done, value } = await reader.read()
if (done) break
buffer += decoder.decode(value, { stream: true })
const lines = buffer.split('\n')
buffer = lines.pop() ?? ''
for (const line of lines) {
const trimmed = line.trim()
if (trimmed) onLine(trimmed)
}
}
if (buffer.trim()) onLine(buffer.trim())
}
}
let runtime: PodmanRuntime | null = null
export function configurePodmanRuntime(config: {
resourcesDir?: string
podmanPath?: string
}): PodmanRuntime {
const podmanPath =
config.podmanPath ??
resolveBundledPodmanPath(config.resourcesDir) ??
'podman'
runtime = new PodmanRuntime({ podmanPath })
return runtime
}
export function getPodmanRuntime(): PodmanRuntime {
if (!runtime) runtime = new PodmanRuntime()
return runtime
}

View File

@@ -1,200 +0,0 @@
import {
type BROWSEROS_ROLE_TEMPLATES,
getBrowserOSRoleTemplate,
} from '@browseros/shared/constants/role-aware-agents'
import type {
BrowserOSAgentRoleId,
BrowserOSAgentRoleSummary,
BrowserOSCustomRoleInput,
BrowserOSRoleTemplate,
} from '@browseros/shared/types/role-aware-agents'
type RoleTemplate = (typeof BROWSEROS_ROLE_TEMPLATES)[number]
interface BootstrapRenderableRole {
name: string
shortDescription: string
longDescription: string
recommendedApps: string[]
boundaries: BrowserOSRoleTemplate['boundaries']
bootstrap: BrowserOSRoleTemplate['bootstrap']
}
export interface RoleBootstrapFiles {
'AGENTS.md': string
'SOUL.md': string
'TOOLS.md': string
'.browseros-role.json': string
}
export function resolveRoleTemplate(
roleId: BrowserOSAgentRoleId,
): RoleTemplate {
const role = getBrowserOSRoleTemplate(roleId)
if (!role) {
throw new Error(`Unknown BrowserOS role: ${roleId}`)
}
return role
}
export function buildRoleBootstrapFiles(input: {
role: BrowserOSRoleTemplate | BrowserOSCustomRoleInput
agentName: string
}): RoleBootstrapFiles {
const normalizedRole = normalizeRoleForBootstrap(input.role)
const roleId = 'id' in input.role ? input.role.id : undefined
return {
'AGENTS.md': normalizedRole.bootstrap.agentsMd,
'SOUL.md': normalizedRole.bootstrap.soulMd,
'TOOLS.md': normalizedRole.bootstrap.toolsMd,
'.browseros-role.json': `${JSON.stringify(
{
version: 1,
roleSource: roleId ? 'builtin' : 'custom',
roleId,
roleName: normalizedRole.name,
shortDescription: normalizedRole.shortDescription,
createdBy: 'browseros',
agentName: input.agentName,
},
null,
2,
)}\n`,
}
}
export function toRoleSummary(
role: BrowserOSRoleTemplate | BrowserOSCustomRoleInput,
): BrowserOSAgentRoleSummary {
const normalizedRole = normalizeRoleForBootstrap(role)
return {
roleSource: 'id' in role ? 'builtin' : 'custom',
roleId: 'id' in role ? role.id : undefined,
roleName: normalizedRole.name,
shortDescription: normalizedRole.shortDescription,
}
}
export function normalizeCustomRole(
role: BrowserOSCustomRoleInput,
): BootstrapRenderableRole {
const recommendedApps = Array.isArray(role.recommendedApps)
? role.recommendedApps.filter(
(app): app is string => typeof app === 'string',
)
: []
const boundaries = Array.isArray(role.boundaries) ? role.boundaries : []
return {
name: role.name,
shortDescription: role.shortDescription,
longDescription: role.longDescription,
recommendedApps,
boundaries,
bootstrap: {
agentsMd:
role.bootstrap?.agentsMd?.trim() ||
buildAgentsMd({
name: role.name,
longDescription: role.longDescription,
boundaries,
}),
soulMd:
role.bootstrap?.soulMd?.trim() ||
buildSoulMd({
name: role.name,
shortDescription: role.shortDescription,
longDescription: role.longDescription,
}),
toolsMd:
role.bootstrap?.toolsMd?.trim() ||
buildToolsMd({
boundaries,
recommendedApps,
}),
},
}
}
function normalizeRoleForBootstrap(
role: BrowserOSRoleTemplate | BrowserOSCustomRoleInput,
): BootstrapRenderableRole {
return 'id' in role ? role : normalizeCustomRole(role)
}
function buildAgentsMd(input: {
name: string
longDescription: string
boundaries: BrowserOSRoleTemplate['boundaries']
}): string {
const boundaryLines = input.boundaries
.map(
(boundary) =>
`- ${boundary.label}: ${boundary.description} Default mode: ${boundary.defaultMode}.`,
)
.join('\n')
return `# ${input.name}
You are the ${input.name} specialist for this workspace.
## Core Purpose
${input.longDescription}
## Operating Rules
${boundaryLines}
## Default Output Style
- concise
- action-oriented
- explicit about blockers and approvals
`
}
function buildSoulMd(input: {
name: string
shortDescription: string
longDescription: string
}): string {
return `# Operating Style
You act like a trusted ${input.name}.
## Working Posture
- calm
- structured
- direct
- explicit about tradeoffs
## Role Framing
${input.shortDescription}
${input.longDescription}
`
}
function buildToolsMd(input: {
boundaries: BrowserOSRoleTemplate['boundaries']
recommendedApps: string[]
}): string {
const boundaryLines = input.boundaries
.map((boundary) => `- ${boundary.label}: ${boundary.defaultMode}`)
.join('\n')
const appsLine =
input.recommendedApps.length > 0
? input.recommendedApps.join(', ')
: 'No specific apps configured yet.'
return `# Tooling Guidelines
- Use BrowserOS MCP for browser and connected SaaS tasks.
- Prefer read, summarize, and draft flows.
- Keep outputs in the workspace when possible so work remains inspectable.
## Recommended Apps
${appsLine}
## Boundary Defaults
${boundaryLines}
`
}

View File

@@ -0,0 +1,114 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Runtime state for the OpenClaw gateway. Today this is just the host port
* we mapped the gateway container to, persisted so that a once-chosen port
* is reused across restarts when it's still free.
*/
import { existsSync } from 'node:fs'
import { mkdir, readFile, writeFile } from 'node:fs/promises'
import { createServer } from 'node:net'
import { join } from 'node:path'
import { OPENCLAW_GATEWAY_CONTAINER_PORT } from '@browseros/shared/constants/openclaw'
import { getOpenClawStateDir } from './openclaw-env'
const RUNTIME_STATE_FILE = 'runtime-state.json'
interface RuntimeState {
gatewayPort: number
}
function readForcedGatewayPort(): number | null {
const raw = process.env.BROWSEROS_TEST_OPENCLAW_GATEWAY_PORT?.trim()
if (!raw) return null
const parsed = Number.parseInt(raw, 10)
if (!Number.isInteger(parsed) || parsed <= 0 || parsed > 65535) {
return null
}
return parsed
}
function getRuntimeStatePath(openclawDir: string): string {
return join(getOpenClawStateDir(openclawDir), RUNTIME_STATE_FILE)
}
export async function readPersistedGatewayPort(
openclawDir: string,
): Promise<number | null> {
const path = getRuntimeStatePath(openclawDir)
if (!existsSync(path)) return null
try {
const parsed = JSON.parse(
await readFile(path, 'utf-8'),
) as Partial<RuntimeState>
if (
typeof parsed.gatewayPort === 'number' &&
Number.isInteger(parsed.gatewayPort) &&
parsed.gatewayPort > 0 &&
parsed.gatewayPort <= 65535
) {
return parsed.gatewayPort
}
return null
} catch {
return null
}
}
async function writePersistedGatewayPort(
openclawDir: string,
port: number,
): Promise<void> {
await mkdir(getOpenClawStateDir(openclawDir), { recursive: true })
const state: RuntimeState = { gatewayPort: port }
await writeFile(
getRuntimeStatePath(openclawDir),
`${JSON.stringify(state, null, 2)}\n`,
)
}
function isPortAvailable(port: number): Promise<boolean> {
return new Promise((resolve) => {
const server = createServer()
server.once('error', () => resolve(false))
server.once('listening', () => {
server.close(() => resolve(true))
})
server.listen(port, '127.0.0.1')
})
}
async function findAvailablePort(startPort: number): Promise<number> {
let port = startPort
while (!(await isPortAvailable(port))) {
port++
}
return port
}
/**
* Pick a host port for the gateway container and persist it. Prefers the
* previously persisted port when it's still bindable; otherwise scans
* upward from OPENCLAW_GATEWAY_CONTAINER_PORT until a free port is found.
*/
export async function allocateGatewayPort(
openclawDir: string,
): Promise<number> {
const forcedPort = readForcedGatewayPort()
if (forcedPort !== null) {
await writePersistedGatewayPort(openclawDir, forcedPort)
return forcedPort
}
const persisted = await readPersistedGatewayPort(openclawDir)
if (persisted !== null && (await isPortAvailable(persisted))) {
return persisted
}
const port = await findAvailablePort(OPENCLAW_GATEWAY_CONTAINER_PORT)
await writePersistedGatewayPort(openclawDir, port)
return port
}

View File

@@ -0,0 +1,61 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { getOpenClawService } from '../openclaw/openclaw-service'
import { OutboundQueueService } from './outbound-queue-service'
let service: OutboundQueueService | null = null
/**
* Lazy singleton — built on first access so the OpenClaw service is
* already available. The queue subscribes to ClawSession state changes
* via OpenClawService.onAgentStatusChange and dispatches through
* OpenClawService.chatStream, so no extra wiring on the openclaw side.
*/
export function getOutboundQueueService(): OutboundQueueService {
if (!service) {
const openclaw = getOpenClawService()
service = new OutboundQueueService({
onAgentStatusChange: (listener) => openclaw.onAgentStatusChange(listener),
getAgentState: (agentId) => openclaw.getAgentState(agentId),
// Resolve the agent's existing user-chat session for queued sends
// so we don't accidentally orphan the conversation by spawning a
// fresh session per queued message. Only the very first message
// for an agent (no prior session at all) falls back to a new key,
// which mirrors what the existing /chat route does.
resolveExistingSessionKey: (agentId) =>
openclaw.resolveAgentSession(agentId).sessionKey ?? null,
chatStream: ({
agentId,
sessionKey,
message,
history,
messageParts,
signal,
}) =>
openclaw.chatStream(agentId, sessionKey, message, history, {
messageParts,
signal,
}),
})
}
return service
}
/** Tear down the singleton — wired into server shutdown. */
export function shutdownOutboundQueueService(): void {
if (service) {
service.shutdown()
service = null
}
}
export type {
QueuedItem,
QueuedItemAttachmentPreview,
QueuedItemPublic,
QueuedItemStatus,
} from './outbound-queue-service'

View File

@@ -0,0 +1,289 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*
* Per-agent FIFO queue of outbound chat messages. The user submits a
* message via /claw/agents/:id/queue, the server holds it, and a worker
* dispatches it through the existing chatStream path the moment the
* agent's ClawSession status flips to idle.
*
* The queue lives in memory only — server restart loses pending items.
* Persistence is a follow-up; the deliberate v1 trade-off is keeping the
* dispatch reactive (single source of truth = ClawSession) and avoiding
* a parallel store that could drift from the agent's actual state.
*/
import { randomUUID } from 'node:crypto'
import { logger } from '../../../lib/logger'
import type {
AgentSessionState,
SessionStateListener,
} from '../openclaw/claw-session'
import type { OpenClawChatContentPart } from '../openclaw/openclaw-http-client'
import type { OpenClawStreamEvent } from '../openclaw/openclaw-types'
export type QueuedItemStatus = 'queued' | 'dispatching' | 'failed'
export interface QueuedItemAttachmentPreview {
kind: 'image' | 'file'
mediaType: string
name?: string
}
export interface QueuedItem {
id: string
agentId: string
/** Plain text body — what we send through chatStream's `message` arg. */
message: string
/** Multimodal parts when attachments are present. */
messageParts?: OpenClawChatContentPart[]
/** Compact preview the SSE feed broadcasts; never includes data URLs. */
attachmentsPreview: QueuedItemAttachmentPreview[]
sessionKey?: string
history: Array<{ role: 'user' | 'assistant'; content: string }>
status: QueuedItemStatus
error?: string
createdAt: number
startedAt?: number
}
/** Public projection sent over the SSE feed — strips heavy fields. */
export interface QueuedItemPublic {
id: string
status: QueuedItemStatus
message: string
attachmentsPreview: QueuedItemAttachmentPreview[]
error?: string
createdAt: number
startedAt?: number
}
interface QueueListener {
agentId: string
send(items: QueuedItemPublic[]): void
}
/** A "send" delegate — wraps OpenClawService.chatStream to avoid a hard dep. */
export type ChatStreamFn = (input: {
agentId: string
sessionKey: string
message: string
history: QueuedItem['history']
messageParts?: OpenClawChatContentPart[]
signal?: AbortSignal
}) => Promise<ReadableStream<OpenClawStreamEvent>>
interface OutboundQueueServiceDeps {
/** Subscribe to per-agent status transitions from the ClawSession SM. */
onAgentStatusChange(listener: SessionStateListener): () => void
/** Read the current ClawSession state for an agent. */
getAgentState(agentId: string): AgentSessionState
/**
* Look up the agent's existing user-chat sessionKey, if any. The worker
* uses this to keep queued sends on the same conversation thread —
* generating a fresh UUID per queued message would orphan the prior
* conversation by spawning a brand-new session each time.
*/
resolveExistingSessionKey(agentId: string): string | null
/** Send a chat — wraps OpenClawService.chatStream. */
chatStream: ChatStreamFn
}
export class OutboundQueueService {
private readonly queues = new Map<string, QueuedItem[]>()
private readonly listeners = new Set<QueueListener>()
private readonly workerInflight = new Map<string, AbortController>()
private unsubscribe: (() => void) | null = null
constructor(private readonly deps: OutboundQueueServiceDeps) {
this.unsubscribe = deps.onAgentStatusChange((agentId, state) => {
if (state.status === 'idle') void this.tryDispatch(agentId)
})
}
enqueue(
item: Omit<QueuedItem, 'id' | 'status' | 'createdAt'> & { id?: string },
): QueuedItem {
// Caller-supplied ids let the browser keep its optimistic row and the
// server snapshot reconciled on a single key — without that, SSE
// can't dedupe the optimistic entry until the POST response lands
// and the client learns the server-generated UUID.
const list = this.queues.get(item.agentId) ?? []
const id =
item.id && !list.some((existing) => existing.id === item.id)
? item.id
: randomUUID()
const queued: QueuedItem = {
...item,
id,
status: 'queued',
createdAt: Date.now(),
}
list.push(queued)
this.queues.set(item.agentId, list)
this.broadcast(item.agentId)
void this.tryDispatch(item.agentId)
return queued
}
cancel(
agentId: string,
itemId: string,
): { ok: true } | { ok: false; reason: 'not_found' | 'dispatching' } {
const list = this.queues.get(agentId) ?? []
const idx = list.findIndex((i) => i.id === itemId)
if (idx < 0) return { ok: false, reason: 'not_found' }
const target = list[idx]
if (!target) return { ok: false, reason: 'not_found' }
if (target.status === 'dispatching') {
return { ok: false, reason: 'dispatching' }
}
list.splice(idx, 1)
this.queues.set(agentId, list)
this.broadcast(agentId)
return { ok: true }
}
retry(agentId: string, itemId: string): { ok: boolean } {
const list = this.queues.get(agentId) ?? []
const item = list.find((i) => i.id === itemId)
if (!item || item.status !== 'failed') return { ok: false }
item.status = 'queued'
item.error = undefined
this.broadcast(agentId)
void this.tryDispatch(agentId)
return { ok: true }
}
list(agentId: string): QueuedItemPublic[] {
const items = this.queues.get(agentId) ?? []
return items.map(toPublic)
}
/** Subscribe to per-agent queue state. Sends a snapshot immediately. */
subscribe(
agentId: string,
send: (items: QueuedItemPublic[]) => void,
): () => void {
const listener: QueueListener = { agentId, send }
this.listeners.add(listener)
try {
send(this.list(agentId))
} catch {
// best effort
}
return () => {
this.listeners.delete(listener)
}
}
private broadcast(agentId: string): void {
const snapshot = this.list(agentId)
for (const listener of this.listeners) {
if (listener.agentId !== agentId) continue
try {
listener.send(snapshot)
} catch {
// ignore — broken listeners GC themselves on next subscribe attempt
}
}
}
private async tryDispatch(agentId: string): Promise<void> {
if (this.workerInflight.has(agentId)) return
const list = this.queues.get(agentId) ?? []
const head = list.find((i) => i.status === 'queued')
if (!head) return
// Don't fire if the agent isn't actually idle yet — even if the
// listener happened to call us early during a state transition.
const state = this.deps.getAgentState(agentId)
if (state.status === 'working') return
head.status = 'dispatching'
head.startedAt = Date.now()
this.broadcast(agentId)
const abort = new AbortController()
this.workerInflight.set(agentId, abort)
try {
// Resolution order: explicit sessionKey on the queued item ➜
// the agent's existing user-chat session ➜ a fresh UUID for the
// first-ever message. This prevents the queue from inadvertently
// splintering an active conversation into a new session.
const targetSessionKey =
head.sessionKey ??
this.deps.resolveExistingSessionKey(agentId) ??
randomUUID()
const stream = await this.deps.chatStream({
agentId,
sessionKey: targetSessionKey,
message: head.message,
history: head.history,
messageParts: head.messageParts,
signal: abort.signal,
})
// Drain the stream to completion so the gateway run finalizes
// properly (writes the JSONL turn, releases the run controller).
const reader = stream.getReader()
try {
while (true) {
if (abort.signal.aborted) break
const { done } = await reader.read()
if (done) break
}
} finally {
await reader.cancel().catch(() => {})
}
this.removeAndBroadcast(agentId, head.id)
} catch (err) {
const message = err instanceof Error ? err.message : String(err)
logger.warn('OutboundQueue dispatch failed', {
agentId,
itemId: head.id,
error: message,
})
head.status = 'failed'
head.error = message
this.broadcast(agentId)
} finally {
this.workerInflight.delete(agentId)
}
// If anything else is still queued and the agent's still idle, drain
// it now without waiting for the next state-change callback.
void this.tryDispatch(agentId)
}
private removeAndBroadcast(agentId: string, itemId: string): void {
const list = this.queues.get(agentId) ?? []
this.queues.set(
agentId,
list.filter((i) => i.id !== itemId),
)
this.broadcast(agentId)
}
shutdown(): void {
this.unsubscribe?.()
this.unsubscribe = null
for (const abort of this.workerInflight.values()) abort.abort()
this.workerInflight.clear()
this.listeners.clear()
this.queues.clear()
}
}
function toPublic(item: QueuedItem): QueuedItemPublic {
return {
id: item.id,
status: item.status,
message: item.message,
attachmentsPreview: item.attachmentsPreview,
error: item.error,
createdAt: item.createdAt,
startedAt: item.startedAt,
}
}

View File

@@ -2,6 +2,7 @@ import {
OPENCLAW_CONTAINER_HOME,
OPENCLAW_TERMINAL_SHELL,
} from '@browseros/shared/constants/openclaw'
import { buildNerdctlCommand } from '../../../lib/container'
import { logger } from '../../../lib/logger'
export const TERMINAL_HOME_DIR = OPENCLAW_CONTAINER_HOME
@@ -11,7 +12,9 @@ const TERMINAL_NAME = 'xterm-256color'
interface TerminalSessionDeps {
containerName: string
podmanPath: string
limaHome: string
limactlPath: string
vmName: string
workingDir: string
onExit: (exitCode: number) => void
onOutput: (data: string) => void
@@ -24,32 +27,44 @@ export interface TerminalSession {
}
export function buildTerminalExecCommand(
podmanPath: string,
limactlPath: string,
vmName: string,
containerName: string,
workingDir: string,
): string[] {
return [
podmanPath,
'exec',
'-it',
'-w',
workingDir,
containerName,
OPENCLAW_TERMINAL_SHELL,
limactlPath,
'shell',
vmName,
'--',
...buildNerdctlCommand([
'exec',
'-it',
'-w',
workingDir,
containerName,
OPENCLAW_TERMINAL_SHELL,
]),
]
}
export function buildTerminalEnv(limaHome: string): NodeJS.ProcessEnv {
return { ...process.env, LIMA_HOME: limaHome, TERM: TERMINAL_NAME }
}
export function createTerminalSession(
deps: TerminalSessionDeps,
): TerminalSession {
const decoder = new TextDecoder()
const proc = Bun.spawn(
buildTerminalExecCommand(
deps.podmanPath,
deps.limactlPath,
deps.vmName,
deps.containerName,
deps.workingDir,
),
{
cwd: '/',
terminal: {
cols: DEFAULT_COLS,
rows: DEFAULT_ROWS,
@@ -58,7 +73,7 @@ export function createTerminalSession(
if (chunk) deps.onOutput(chunk)
},
},
env: { ...process.env, TERM: TERMINAL_NAME },
env: buildTerminalEnv(deps.limaHome),
},
)
let closed = false

View File

@@ -517,15 +517,45 @@ export class Browser {
return null
}
private async resolveWindowIdForNewPage(opts?: {
hidden?: boolean
windowId?: number
}): Promise<number | undefined> {
if (!opts?.hidden) {
return opts?.windowId
}
if (opts.windowId !== undefined) {
const windows = await this.listWindows()
const targetWindow = windows.find(
(window) => window.windowId === opts.windowId,
)
if (targetWindow && !targetWindow.isVisible) {
return targetWindow.windowId
}
if (targetWindow?.isVisible) {
logger.warn(
'Requested hidden page target window is visible, creating a new hidden window instead',
{
requestedWindowId: opts.windowId,
},
)
}
}
const hiddenWindow = await this.createWindow({ hidden: true })
return hiddenWindow.windowId
}
async newPage(
url: string,
opts?: { hidden?: boolean; background?: boolean; windowId?: number },
): Promise<number> {
const windowId = await this.resolveWindowIdForNewPage(opts)
const createResult = await this.cdp.Browser.createTab({
url,
...(opts?.hidden !== undefined && { hidden: opts.hidden }),
...(opts?.background !== undefined && { background: opts.background }),
...(opts?.windowId !== undefined && { windowId: opts.windowId }),
...(windowId !== undefined && { windowId }),
})
const tabId = (createResult.tab as TabInfo).tabId
@@ -553,7 +583,7 @@ export class Browser {
loadProgress: tabInfo.loadProgress,
isPinned: tabInfo.isPinned,
isHidden: tabInfo.isHidden,
windowId: tabInfo.windowId,
windowId: tabInfo.windowId ?? windowId,
index: tabInfo.index,
groupId: tabInfo.groupId,
})

View File

@@ -8,6 +8,7 @@
import fs from 'node:fs'
import path from 'node:path'
import { EXTERNAL_URLS } from '@browseros/shared/constants/urls'
import { Command, InvalidArgumentError } from 'commander'
import { z } from 'zod'
@@ -30,6 +31,8 @@ export const ServerConfigSchema = z.object({
instanceBrowserosVersion: z.string().optional(),
instanceChromiumVersion: z.string().optional(),
aiSdkDevtoolsEnabled: z.boolean(),
vmCachePrefetch: z.boolean(),
vmCacheManifestUrl: z.string().url(),
})
export type ServerConfig = z.infer<typeof ServerConfigSchema>
@@ -226,6 +229,11 @@ function parseConfigFile(filePath?: string): ConfigResult<PartialConfig> {
cfg.flags?.allow_remote_in_mcp === true ? true : undefined,
aiSdkDevtoolsEnabled:
cfg.flags?.ai_sdk_devtools === true ? true : undefined,
vmCachePrefetch:
typeof cfg.vm_cache?.prefetch === 'boolean'
? cfg.vm_cache.prefetch
: undefined,
vmCacheManifestUrl: parseTrimmedString(cfg.vm_cache?.manifest_url),
instanceClientId:
typeof cfg.instance?.client_id === 'string'
? cfg.instance.client_id
@@ -272,6 +280,10 @@ function parseRuntimeEnv(): PartialConfig {
instanceClientId: process.env.BROWSEROS_CLIENT_ID,
aiSdkDevtoolsEnabled:
process.env.BROWSEROS_AI_SDK_DEVTOOLS === 'true' ? true : undefined,
vmCachePrefetch: parseBooleanEnv(process.env.BROWSEROS_VM_CACHE_PREFETCH),
vmCacheManifestUrl: parseTrimmedString(
process.env.BROWSEROS_VM_CACHE_MANIFEST_URL,
),
})
}
@@ -305,6 +317,8 @@ function getDefaults(cwd: string): PartialConfig {
executionDir: cwd,
mcpAllowRemote: false,
aiSdkDevtoolsEnabled: false,
vmCachePrefetch: true,
vmCacheManifestUrl: EXTERNAL_URLS.VM_CACHE_MANIFEST,
}
}
@@ -325,6 +339,18 @@ function safeParseInt(value: string): number | undefined {
return Number.isNaN(num) ? undefined : num
}
function parseBooleanEnv(value: string | undefined): boolean | undefined {
if (value === 'true') return true
if (value === 'false') return false
return undefined
}
function parseTrimmedString(value: unknown): string | undefined {
if (typeof value !== 'string') return undefined
const trimmed = value.trim()
return trimmed.length > 0 ? trimmed : undefined
}
function omitUndefined<T extends Record<string, unknown>>(obj: T): Partial<T> {
return Object.fromEntries(
Object.entries(obj).filter(([_, v]) => v !== undefined),

View File

@@ -19,6 +19,8 @@ export const INLINED_ENV = {
CODEGEN_SERVICE_URL: process.env.CODEGEN_SERVICE_URL,
POSTHOG_API_KEY: process.env.POSTHOG_API_KEY,
BROWSEROS_CONFIG_URL: process.env.BROWSEROS_CONFIG_URL,
BROWSEROS_VM_CACHE_PREFETCH: process.env.BROWSEROS_VM_CACHE_PREFETCH,
BROWSEROS_VM_CACHE_MANIFEST_URL: process.env.BROWSEROS_VM_CACHE_MANIFEST_URL,
SKILLS_CATALOG_URL: process.env.SKILLS_CATALOG_URL,
} as const
@@ -27,4 +29,6 @@ export const REQUIRED_FOR_PRODUCTION = [
'CODEGEN_SERVICE_URL',
'POSTHOG_API_KEY',
'BROWSEROS_CONFIG_URL',
'BROWSEROS_VM_CACHE_PREFETCH',
'BROWSEROS_VM_CACHE_MANIFEST_URL',
] as const satisfies readonly (keyof typeof INLINED_ENV)[]

View File

@@ -7,7 +7,20 @@ import type { ServerDiscoveryConfig } from '@browseros/shared/types/server-confi
import { logger } from './logger'
export function getBrowserosDir(): string {
return join(homedir(), PATHS.BROWSEROS_DIR_NAME)
const override = process.env.BROWSEROS_DIR?.trim()
if (override) {
return override
}
const dirName =
process.env.NODE_ENV === 'development'
? PATHS.DEV_BROWSEROS_DIR_NAME
: PATHS.BROWSEROS_DIR_NAME
return join(homedir(), dirName)
}
export function logDevelopmentBrowserosDir(): void {
if (process.env.NODE_ENV !== 'development') return
logger.info(`Using development BrowserOS directory: ${getBrowserosDir()}`)
}
export function getMemoryDir(): string {
@@ -35,9 +48,49 @@ export function getBuiltinSkillsDir(): string {
}
export function getOpenClawDir(): string {
return join(getVmStateDir(), PATHS.OPENCLAW_DIR_NAME)
}
export function getLegacyOpenClawDir(): string {
return join(getBrowserosDir(), PATHS.OPENCLAW_DIR_NAME)
}
export function getCacheDir(): string {
return join(getBrowserosDir(), PATHS.CACHE_DIR_NAME)
}
export function getVmCacheDir(): string {
return join(getCacheDir(), 'vm')
}
export function getLimaHomeDir(): string {
return join(getBrowserosDir(), 'lima')
}
export function getVmStateDir(): string {
return join(getBrowserosDir(), 'vm')
}
export function getVmDisksDir(): string {
return getVmCacheDir()
}
export function getAgentCacheDir(): string {
return join(getVmCacheDir(), 'images')
}
export function getLazyMonitoringDir(): string {
return join(getBrowserosDir(), 'lazy-monitoring')
}
export function getLazyMonitoringRunsDir(): string {
return join(getLazyMonitoringDir(), 'runs')
}
export function getLazyMonitoringRunDir(runId: string): string {
return join(getLazyMonitoringRunsDir(), runId)
}
export function getServerConfigPath(): string {
return join(getBrowserosDir(), PATHS.SERVER_CONFIG_FILE_NAME)
}
@@ -57,10 +110,13 @@ export function removeServerConfigSync(): void {
}
export async function ensureBrowserosDir(): Promise<void> {
logDevelopmentBrowserosDir()
await mkdir(getMemoryDir(), { recursive: true })
await mkdir(getSkillsDir(), { recursive: true })
await mkdir(getBuiltinSkillsDir(), { recursive: true })
await mkdir(getSessionsDir(), { recursive: true })
await mkdir(getLazyMonitoringRunsDir(), { recursive: true })
await mkdir(getAgentCacheDir(), { recursive: true })
}
export async function cleanOldSessions(): Promise<void> {

View File

@@ -0,0 +1,209 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { ContainerCliError } from '../vm/errors'
import { LimaCli } from '../vm/lima-cli'
import type { ContainerSpec, LogFn, MountSpec, PortMapping } from './types'
export function buildNerdctlCommand(args: string[]): string[] {
return ['nerdctl', ...args]
}
export interface ContainerCliConfig {
limactlPath: string
limaHome: string
vmName: string
sshPath?: string
}
export interface ContainerCommandResult {
exitCode: number
stdout: string
stderr: string
}
export class ContainerCli {
private readonly lima: LimaCli
constructor(private readonly cfg: ContainerCliConfig) {
this.lima = new LimaCli({
limactlPath: cfg.limactlPath,
limaHome: cfg.limaHome,
sshPath: cfg.sshPath,
})
}
async imageExists(ref: string): Promise<boolean> {
const result = await this.runCommand(['image', 'inspect', ref])
return result.exitCode === 0
}
async pullImage(ref: string, onLog?: LogFn): Promise<void> {
await this.runRequired(['pull', ref], onLog)
}
async loadImage(tarballPath: string, onLog?: LogFn): Promise<string[]> {
const result = await this.runRequired(['load', '-i', tarballPath], onLog)
return parseLoadedImageRefs(result.stdout)
}
async createContainer(spec: ContainerSpec, onLog?: LogFn): Promise<void> {
await this.runRequired(buildCreateArgs(spec), onLog)
}
async startContainer(name: string, onLog?: LogFn): Promise<void> {
await this.runRequired(['start', name], onLog)
}
async stopContainer(name: string, onLog?: LogFn): Promise<void> {
const result = await this.runCommand(['stop', name], onLog)
if (result.exitCode === 0 || isNoSuchContainer(result.stderr)) return
throw this.commandError(['stop', name], result)
}
async removeContainer(
name: string,
opts?: { force?: boolean },
onLog?: LogFn,
): Promise<void> {
const args = ['rm']
if (opts?.force) args.push('-f')
args.push(name)
const result = await this.runCommand(args, onLog)
if (result.exitCode === 0 || isNoSuchContainer(result.stderr)) return
throw this.commandError(args, result)
}
async exec(name: string, cmd: string[], onLog?: LogFn): Promise<number> {
const result = await this.runCommand(['exec', name, ...cmd], onLog)
return result.exitCode
}
async ps(opts?: { namesOnly?: boolean }): Promise<string[]> {
const args = opts?.namesOnly ? ['ps', '--format', '{{.Names}}'] : ['ps']
const result = await this.runRequired(args)
return result.stdout
.trim()
.split('\n')
.map((line) => line.trim())
.filter(Boolean)
}
tailLogs(name: string, onLine: LogFn): () => void {
const proc = this.lima.spawnShell(
this.cfg.vmName,
buildNerdctlCommand(['logs', '-f', '-n', '0', name]),
{ onStdout: onLine, onStderr: onLine },
)
let stopped = false
return () => {
if (stopped) return
stopped = true
proc.kill()
}
}
async runCommand(
args: string[],
onLog?: LogFn,
): Promise<ContainerCommandResult> {
const stdoutLines: string[] = []
const stderrLines: string[] = []
const exitCode = await this.lima.shell(
this.cfg.vmName,
buildNerdctlCommand(args),
{
onStdout: (line) => {
stdoutLines.push(line)
onLog?.(line)
},
onStderr: (line) => {
stderrLines.push(line)
onLog?.(line)
},
},
)
return {
exitCode,
stdout: linesToOutput(stdoutLines),
stderr: stderrLines.join('\n'),
}
}
private async runRequired(
args: string[],
onLog?: LogFn,
): Promise<ContainerCommandResult> {
const result = await this.runCommand(args, onLog)
if (result.exitCode === 0) return result
throw this.commandError(args, result)
}
private commandError(
args: string[],
result: ContainerCommandResult,
): ContainerCliError {
return new ContainerCliError(
`nerdctl ${args.join(' ')}`,
result.exitCode,
result.stderr.trim(),
)
}
}
function buildCreateArgs(spec: ContainerSpec): string[] {
const args = ['create', '--name', spec.name]
if (spec.restart) args.push('--restart', spec.restart)
for (const port of spec.ports ?? []) args.push('-p', portArg(port))
if (spec.envFile) args.push('--env-file', spec.envFile)
for (const [key, value] of Object.entries(spec.env ?? {})) {
args.push('-e', `${key}=${value}`)
}
for (const mount of spec.mounts ?? []) args.push('-v', mountArg(mount))
for (const host of spec.addHosts ?? []) args.push('--add-host', host)
if (spec.health) {
args.push('--health-cmd', spec.health.cmd)
if (spec.health.interval)
args.push('--health-interval', spec.health.interval)
if (spec.health.timeout) args.push('--health-timeout', spec.health.timeout)
if (spec.health.retries !== undefined) {
args.push('--health-retries', String(spec.health.retries))
}
}
args.push(spec.image)
args.push(...(spec.command ?? []))
return args
}
function portArg(port: PortMapping): string {
const host = port.hostIp ? `${port.hostIp}:${port.hostPort}` : port.hostPort
return `${host}:${port.containerPort}`
}
function mountArg(mount: MountSpec): string {
return `${mount.source}:${mount.target}${mount.readonly ? ':ro' : ''}`
}
function parseLoadedImageRefs(stdout: string): string[] {
return stdout
.split('\n')
.map((line) => line.match(/^Loaded image(?:\(s\))?:\s*(.+)$/i)?.[1]?.trim())
.filter((ref): ref is string => !!ref)
}
function isNoSuchContainer(stderr: string): boolean {
const lower = stderr.toLowerCase()
return lower.includes('no such container') || lower.includes('not found')
}
function linesToOutput(lines: string[]): string {
if (lines.length === 0) return ''
return `${lines.join('\n')}\n`
}

View File

@@ -0,0 +1,64 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { basename, join } from 'node:path'
import { ContainerCliError, ImageLoadError } from '../vm/errors'
import type { VmManifest } from '../vm/manifest'
import type { Arch } from '../vm/paths'
import { getImageCacheDir, hostPathToGuest } from '../vm/paths'
import type { ContainerCli } from './container-cli'
import type { LogFn } from './types'
export class ImageLoader {
constructor(
private readonly cli: ContainerCli,
private readonly manifest: VmManifest,
private readonly arch: Arch,
private readonly browserosRoot?: string,
) {}
async ensureImageLoaded(ref: string, onLog?: LogFn): Promise<void> {
if (await this.cli.imageExists(ref)) return
const tarball = this.resolveTarball(ref)
const hostPath = join(
getImageCacheDir(this.browserosRoot),
basename(tarball.key),
)
const guestPath = hostPathToGuest(hostPath, this.browserosRoot)
try {
await this.cli.loadImage(guestPath, onLog)
} catch (error) {
if (error instanceof ContainerCliError) {
throw new ImageLoadError(ref, `load failed: ${error.stderr}`, error)
}
throw error
}
if (!(await this.cli.imageExists(ref))) {
throw new ImageLoadError(
ref,
`image not present after successful load of ${guestPath}`,
)
}
}
private resolveTarball(
ref: string,
): VmManifest['agents'][string]['tarballs'][Arch] {
for (const agent of Object.values(this.manifest.agents)) {
if (`${agent.image}:${agent.version}` !== ref) continue
const tarball = agent.tarballs[this.arch]
if (!tarball) {
throw new ImageLoadError(ref, `no ${this.arch} tarball in manifest`)
}
return tarball
}
throw new ImageLoadError(ref, `no agent in manifest matches ${ref}`)
}
}

View File

@@ -0,0 +1,9 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
export * from './container-cli'
export * from './image-loader'
export * from './types'

View File

@@ -0,0 +1,44 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
export type LogFn = (msg: string) => void
export interface PortMapping {
hostIp?: string
hostPort: number
containerPort: number
}
export interface MountSpec {
source: string
target: string
readonly?: boolean
}
export interface HealthConfig {
cmd: string
interval?: string
timeout?: string
retries?: number
}
export interface ContainerSpec {
name: string
image: string
restart?: 'no' | 'unless-stopped' | 'always'
ports?: PortMapping[]
env?: Record<string, string>
envFile?: string
mounts?: MountSpec[]
addHosts?: string[]
health?: HealthConfig
command?: string[]
}
export interface LogLine {
stream: 'stdout' | 'stderr'
line: string
}

View File

@@ -0,0 +1,322 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
import { createHash } from 'node:crypto'
import { createReadStream, existsSync } from 'node:fs'
import { mkdir, readFile, rename, rm } from 'node:fs/promises'
import { arch as hostArch } from 'node:os'
import { dirname, join } from 'node:path'
import { EXTERNAL_URLS } from '@browseros/shared/constants/urls'
import type { VmArtifact, VmManifest } from './manifest'
import type { Arch } from './paths'
import { getCachedManifestPath } from './paths'
const DEFAULT_TIMEOUT_MS = 30_000
const ARCHES: Arch[] = ['arm64', 'x64']
const CANONICAL_MANIFEST_SUFFIX = '/vm/manifest.json'
export interface VmCacheSyncOptions {
browserosRoot?: string
manifestUrl?: string
allArches?: boolean
fetchImpl?: typeof fetch
rawHostArch?: NodeJS.Architecture
timeoutMs?: number
}
export interface VmCacheSyncResult {
downloaded: string[]
manifestPath: string
skipped: boolean
}
const inFlight = new Map<string, Promise<VmCacheSyncResult>>()
export function prefetchVmCache(
options: VmCacheSyncOptions = {},
): Promise<VmCacheSyncResult> {
return startOrReuseSync(options)
}
export function ensureVmCacheSynced(
options: VmCacheSyncOptions = {},
): Promise<VmCacheSyncResult> {
return startOrReuseSync(options)
}
export async function ensureVmCacheAvailable(
options: VmCacheSyncOptions = {},
): Promise<void> {
const cfg = resolveSyncConfig(options)
const pending = inFlight.get(syncKey(cfg))
if (pending) {
await pending.catch(() => {})
}
if (existsSync(getCachedManifestPath(cfg.browserosRoot))) return
await startOrReuseSyncWithConfig(cfg)
}
function startOrReuseSync(
options: VmCacheSyncOptions,
): Promise<VmCacheSyncResult> {
try {
return startOrReuseSyncWithConfig(resolveSyncConfig(options))
} catch (error) {
return Promise.reject(error)
}
}
function startOrReuseSyncWithConfig(
cfg: SyncConfig,
): Promise<VmCacheSyncResult> {
const key = syncKey(cfg)
const existing = inFlight.get(key)
if (existing) return existing
const current = syncVmCache(cfg).finally(() => {
if (inFlight.get(key) === current) inFlight.delete(key)
})
inFlight.set(key, current)
return current
}
async function syncVmCache(cfg: SyncConfig): Promise<VmCacheSyncResult> {
const remote = await fetchManifest(cfg)
const manifestPath = getCachedManifestPath(cfg.browserosRoot)
const local = await readLocalManifest(manifestPath)
const plan = await planDownloads({
remote,
local,
cacheRoot: cacheRootForManifest(manifestPath),
arches: cfg.arches,
})
for (const item of plan) {
await downloadArtifact(
cfg.fetchImpl,
artifactUrlForKey(cfg.manifestUrl, item.key),
item.destPath,
item.sha256,
cfg.timeoutMs,
)
}
await mkdir(dirname(manifestPath), { recursive: true })
const tempPath = `${manifestPath}.${process.pid}.${Date.now()}.tmp`
await Bun.write(tempPath, `${JSON.stringify(remote, null, 2)}\n`)
await rename(tempPath, manifestPath)
return {
downloaded: plan.map((item) => item.key),
manifestPath,
skipped: plan.length === 0,
}
}
interface SyncConfig {
browserosRoot?: string
manifestUrl: string
fetchImpl: typeof fetch
arches: Arch[]
timeoutMs: number
}
function resolveSyncConfig(options: VmCacheSyncOptions): SyncConfig {
return {
browserosRoot: options.browserosRoot,
manifestUrl:
trimNonEmpty(options.manifestUrl) ??
trimNonEmpty(process.env.BROWSEROS_VM_CACHE_MANIFEST_URL) ??
EXTERNAL_URLS.VM_CACHE_MANIFEST,
fetchImpl: options.fetchImpl ?? fetch,
arches: selectSyncArches(options),
timeoutMs: options.timeoutMs ?? DEFAULT_TIMEOUT_MS,
}
}
async function fetchManifest(cfg: SyncConfig): Promise<VmManifest> {
const response = await fetchWithTimeout(
cfg.fetchImpl,
cfg.manifestUrl,
cfg.timeoutMs,
)
if (!response.ok) {
throw new Error(
`manifest fetch failed: ${cfg.manifestUrl} (${response.status})`,
)
}
return (await response.json()) as VmManifest
}
interface DownloadPlanItem {
key: string
destPath: string
sha256: string
}
async function planDownloads(opts: {
remote: VmManifest
local: VmManifest | null
cacheRoot: string
arches: Arch[]
}): Promise<DownloadPlanItem[]> {
const out: DownloadPlanItem[] = []
for (const arch of opts.arches) {
for (const [name, agent] of Object.entries(opts.remote.agents)) {
const remote = agent.tarballs[arch]
if (!remote) continue
const destPath = join(opts.cacheRoot, remote.key)
if (
!(await needsDownload(
remote,
opts.local?.agents[name]?.tarballs[arch],
destPath,
))
) {
continue
}
out.push({ key: remote.key, destPath, sha256: remote.sha256 })
}
}
return out
}
async function needsDownload(
remote: VmArtifact,
local: VmArtifact | undefined,
destPath: string,
): Promise<boolean> {
if (!existsSync(destPath)) return true
if (local?.sha256 === remote.sha256) return false
try {
return (await sha256File(destPath)) !== remote.sha256
} catch {
return true
}
}
async function downloadArtifact(
fetchImpl: typeof fetch,
url: string,
destPath: string,
sha256: string,
timeoutMs: number,
): Promise<void> {
const partialPath = `${destPath}.partial`
await mkdir(dirname(destPath), { recursive: true })
await rm(partialPath, { force: true })
try {
const response = await fetchWithTimeout(fetchImpl, url, timeoutMs)
if (!response.ok || !response.body) {
throw new Error(`download failed: ${url} (${response.status})`)
}
const sink = Bun.file(partialPath).writer()
const reader = response.body.getReader()
try {
for (;;) {
const { done, value } = await reader.read()
if (done) break
sink.write(value)
}
} finally {
await sink.end()
}
await verifySha256(partialPath, sha256)
await rename(partialPath, destPath)
} catch (error) {
await rm(partialPath, { force: true })
throw error
}
}
async function fetchWithTimeout(
fetchImpl: typeof fetch,
url: string,
timeoutMs: number,
): Promise<Response> {
const controller = new AbortController()
const timer = setTimeout(() => controller.abort(), timeoutMs)
try {
return await fetchImpl(url, { signal: controller.signal })
} catch (error) {
if ((error as { name?: string }).name === 'AbortError') {
throw new Error(`fetch timed out after ${timeoutMs}ms: ${url}`)
}
throw error
} finally {
clearTimeout(timer)
}
}
async function verifySha256(path: string, expected: string): Promise<void> {
const actual = await sha256File(path)
if (actual !== expected) {
throw new Error(
`sha256 mismatch for ${path}: expected ${expected}, got ${actual}`,
)
}
}
async function sha256File(path: string): Promise<string> {
const hash = createHash('sha256')
for await (const chunk of createReadStream(path)) {
hash.update(chunk)
}
return hash.digest('hex')
}
async function readLocalManifest(path: string): Promise<VmManifest | null> {
try {
return JSON.parse(await readFile(path, 'utf8')) as VmManifest
} catch (error) {
if ((error as NodeJS.ErrnoException).code === 'ENOENT') return null
throw error
}
}
function selectSyncArches(options: VmCacheSyncOptions): Arch[] {
if (options.allArches) return [...ARCHES]
const rawArch = options.rawHostArch ?? hostArch()
if (rawArch === 'arm64') return ['arm64']
if (rawArch === 'x64' || rawArch === 'ia32') return ['x64']
throw new Error(`unsupported host arch: ${rawArch}`)
}
function cacheRootForManifest(manifestPath: string): string {
return dirname(dirname(manifestPath))
}
function syncKey(cfg: SyncConfig): string {
return [
getCachedManifestPath(cfg.browserosRoot),
cfg.manifestUrl,
cfg.arches.join(','),
String(cfg.timeoutMs),
].join('\0')
}
function artifactUrlForKey(manifestUrl: string, key: string): string {
const artifactKey = key.replace(/^\/+/, '')
const url = new URL(manifestUrl)
const normalizedPath = url.pathname.replace(/\/+$/, '')
const prefix = normalizedPath.endsWith(CANONICAL_MANIFEST_SUFFIX)
? normalizedPath.slice(0, -CANONICAL_MANIFEST_SUFFIX.length)
: normalizedPath.slice(0, Math.max(0, normalizedPath.lastIndexOf('/')))
url.pathname = `${prefix.replace(/\/+$/, '')}/${artifactKey}`
url.search = ''
url.hash = ''
return url.toString()
}
function trimNonEmpty(value: string | undefined): string | undefined {
const trimmed = value?.trim()
return trimmed ? trimmed : undefined
}

View File

@@ -0,0 +1,60 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
export class VmError extends Error {
constructor(message: string) {
super(message)
this.name = new.target.name
}
}
export class VmNotReadyError extends VmError {}
export class VmStateCorruptedError extends VmError {}
export class LimaCommandError extends VmError {
constructor(
command: string,
public readonly exitCode: number,
public readonly stderr: string,
) {
super(`${command} failed with exit code ${exitCode}: ${stderr}`)
}
}
export class ContainerCliError extends VmError {
constructor(
command: string,
public readonly exitCode: number,
public readonly stderr: string,
) {
super(`${command} failed with exit code ${exitCode}: ${stderr}`)
}
}
export class ImageLoadError extends VmError {
constructor(
public readonly imageRef: string,
message: string,
public override readonly cause?: unknown,
) {
super(`failed to load image ${imageRef}: ${message}`)
}
}
export class ManifestMissingError extends VmError {
constructor(public readonly manifestPath: string) {
super(manifestMissingMessage(manifestPath))
}
}
function manifestMissingMessage(manifestPath: string): string {
const message = `VM manifest is missing at ${manifestPath}`
if (process.env.NODE_ENV === 'development') {
return `${message}; run bun run dev:setup before starting the server`
}
return message
}

View File

@@ -0,0 +1,13 @@
/**
* @license
* Copyright 2025 BrowserOS
* SPDX-License-Identifier: AGPL-3.0-or-later
*/
export * from './errors'
export * from './lima-cli'
export * from './lima-config'
export * from './manifest'
export * from './paths'
export * from './telemetry'
export * from './vm-runtime'

Some files were not shown because too many files have changed in this diff Show More