* fix: compaction config for small context windows (≤32K)
Raise COMPACTION_SMALL_CONTEXT_WINDOW from 16K to 32K so models like
Haiku 4.5 (30K context) use proportional 50% reserve instead of the
fixed 20K reserve. Also scale fixedOverhead for small contexts (capped
at 40% of context window) to prevent the doom loop where overhead alone
triggers compaction on every step.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* docs: add compaction tuning guidance to limits constants
Explain the relationship between SMALL_CONTEXT_WINDOW and
FIXED_OVERHEAD so devs know the 24K minimum constraint when
tweaking these values.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Add window focus listener in ChatFooter that focuses the textarea when
the side panel receives focus. Handles both initial open (via
document.hasFocus check on mount) and re-focus scenarios (via window
focus event). Guards against stealing focus from other interactive
elements.
Companion Chromium fix: side_panel_coordinator.cc now always calls
RequestFocus() in PopulateSidePanel(), not just when there's no
previous entry — ensuring the side panel WebContents receives focus
on every open/toggle.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat: add 2-stage pruning to compaction pipeline before LLM summarization
Add two new lightweight stages to the compaction prepareStep pipeline that
recover context tokens cheaply before falling back to expensive LLM
summarization:
- Stage 2: Use AI SDK's pruneMessages to remove old tool call/result
pairs beyond the last 6 messages entirely
- Stage 3: Replace remaining tool output values with short placeholders
("[Cleared — N chars]") while preserving tool call structure and IDs
Both stages re-estimate tokens from message content (not stale step
usage) after modifying messages. The existing LLM summarization and
sliding window fallback remain as Stage 4.
Also adds estimateTokensForThreshold() helper, clearToolOutputs()
function, and COMPACTION_PRUNE_KEEP_RECENT_MESSAGES /
COMPACTION_CLEAR_OUTPUT_MIN_CHARS constants.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: reorder compaction pipeline — truncate before clear, protect recent tools
- Stage 0: Check threshold, return untouched when under (no data loss)
- Stage 1: Prune old tool call/result pairs beyond last 6 messages
- Stage 2: Truncate large tool outputs to 15K chars (keeps partial content)
- Stage 3: Clear old tool outputs with placeholders, protect last 2
- Stage 4: LLM-based compaction with sliding window fallback
clearToolOutputs now accepts keepRecentCount parameter (default 2) to
skip the N most recent tool messages from clearing.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: limits fixes
* fix: address review — preserve toKeep context, derive test values from constants
- When Stage 3 (clearToolOutputs) doesn't resolve overflow, pass
truncated (not cleared) messages to Stage 4 so toKeep retains
meaningful tool outputs for the agent's immediate context
- Add comment explaining intentional conservatism in post-prune
token estimation (step usage is stale, must re-estimate safely)
- Refactor computeConfig tests to derive expected values from
AGENT_LIMITS constants instead of hardcoding magic numbers
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The system prompt referenced `browser_open_tab` which was renamed to
`new_page`. This caused models to infer a `browser_*` naming convention
and call non-existent tools like `browser_navigate`, resulting in
MCP error -32602.
Fixes TKT-540
Add changelog entry for BrowserOS v0.42.0 featuring SOUL.md, vertical tabs,
long-term memory, and Chromium 146 update. Include screenshots from the
GitHub release.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: new tools for breadcrumbs
* feat: setup scheduled task card
* feat: added dismiss cooldown
* chore: update prompt
* fix: support api key tool
* fix: prompt text to limit nudges
* fix: scheduled tasks card
* fix: update nudges prompt
* feat: skip nudges when user dismisses nudge
* fix: ensure nudges only show if they are not dismissed
* Revert "fix: ensure nudges only show if they are not dismissed"
This reverts commit d825254698829b8e9941aae7873bd440027d0c74.
* Revert "feat: skip nudges when user dismisses nudge"
This reverts commit 12b552b454d10ec4209b88668fc48681423ff6fc.
* Revert "fix: update nudges prompt"
This reverts commit 80b7520b953b4d3cbed2ed477b9e508e39938dca.
* feat: update agent with mcp when new mcp connection is added
* feat: created connect apps option as a blocking card system
* feat: schedule tasks passive without dismiss
* fix: nudges and prompt texts
* fix: biome lint errors
* fix: review comments
* fix: resolve comments
* fix: review comments
* fix: review comments
* fix: auto resolve state
* fix: eliminate the race where the async delete could resolve after the
new session
* feat: track ignored apps list
* fix: empty response text object on message reply
* feat: sync previously connected mcps
* feat: sync integrations with klavis
* feat: account for unauthenticated connections
* fix: analytics events
* fix: typescript issues
* fix: klavis client issue
* fix: invalid mcps causing entire responses from failing
* fix: prompt with card for integrations when the integration fails
* fix: prompt structure to support declined apps
* fix: refresh session on mcp changes
* feat: add agent skills system with catalog, loader, and UI
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: return 500 for server errors in PUT/DELETE skill routes
Previously both handlers returned 404 for all errors, masking filesystem
failures (disk full, permission denied) as "not found". Now only
"not found" errors return 404; everything else returns 500.
* fix: align SKILL.md format with agentskills.io spec
- Move `enabled` and `version` into `metadata` field (spec only allows
name, description, license, compatibility, metadata, allowed-tools)
- Frontmatter `name` now matches directory name (lowercase kebab-case)
- Human-readable name stored in `metadata.display-name`
- Add index signature to SkillMetadata for arbitrary string keys
- Validate frontmatter with type guard in getSkill (remove unsafe cast)
- updateSkill now preserves existing frontmatter fields (license, etc.)
- Tighten buildSkillMd param from Record<string, unknown> to SkillFrontmatter
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- truncateToolOutputs: handle all output.type variants (text, json,
content) by checking output.value directly instead of branching on
type. The old code missed type 'content' (array of content parts),
causing 1M+ char tool results to pass through untouched.
- estimateTokens: change chars/4 to chars/3 — HTML/Markdown content
tokenizes at ~3.14 chars/token empirically, not 4.
- COMPACTION_FIXED_OVERHEAD: 5K → 12K to account for system prompt
(~2.5K tokens) + tool definitions as JSON Schema (~8-9K tokens).
- Apply truncateToolOutputs in prepareStep (Stage 0) before token
estimation, not just during summarization.
* fix: support artifact-extracted directory structure in OTA binary discovery
The download_resources system now extracts server binaries into
platform-specific subdirectories (e.g., darwin-arm64/resources/bin/),
but the OTA module only looked for flat binary names. This adds
find_server_binary() which checks both layouts, keeping backward
compatibility with --binaries while supporting the new structure.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: download server binaries from R2 instead of requiring --binaries
Remove the --binaries flag from `ota server release`. The module now
downloads artifact zips from artifacts/server/latest/ in R2, extracts
them, then signs and packages as before. This eliminates the need to
have mono build output locally.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: robust compaction with Pi-style token counting + overflow middleware
Root cause: getCurrentTokenCount() returned stale inputTokens from the
previous step, ignoring new tool results added to messages since that
step. A large tool output (DOM snapshot, page content) caused a token
jump that bypassed the compaction threshold check, leading to
context_length_exceeded errors (322K tokens sent, model max 262K).
Layer 1 — Accurate token counting (proactive):
- Adopt Pi coding agent's additive approach: base(inputTokens) +
outputTokens + estimate(trailing tool results)
- Trailing tool results are estimated by walking backwards from end of
messages array until a non-tool message is found
- Falls back to full estimation with safety multiplier when no real
usage data is available (first step of a turn)
Layer 2 — Context overflow middleware (reactive):
- LanguageModelV3Middleware that wraps doGenerate/doStream
- Catches context_length_exceeded errors at the model call level
- Truncates prompt (keeps system messages + most recent non-system
messages targeting 60% of context window)
- Retries the model call once
Verified end-to-end with real model (Gemini Flash Lite via OpenRouter)
on 16K context window: 4 compactions triggered correctly across 8
steps, no context_length_exceeded errors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: adopt Pi-style overflow detection patterns + fix truncation edge case
- Replace 6 generic substring matches with 17 provider-specific regex
patterns from Pi coding agent (Anthropic, OpenAI, Google, xAI, Groq,
OpenRouter, Bedrock, Copilot, llama.cpp, LM Studio, MiniMax, Kimi,
Mistral, z.ai)
- Fix truncatePrompt edge case: when the last message alone exceeds the
target, keepFrom was never updated → empty non-system messages. Now
always keeps at least the most recent non-system message.
- Add runtime guard for LanguageModelV3 cast in ai-sdk-agent.ts
- Add tests for false-positive rejection and truncation edge case
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The Kimi K2.5 model supports a 256,000 token context window, not
128,000. Updated the provider template and model config to reflect
the correct value.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat: return element coordinates in tool responses and DPR in screenshots
- click, hover, fill, drag now return resolved coordinates in response text
- take_screenshot returns devicePixelRatio for mapping coordinates to pixels
- Coordinates are in CSS pixels; multiply by DPR to get screenshot pixels
* fix: use Promise.allSettled in screenshot to prevent DPR eval from aborting capture
Runtime.evaluate for devicePixelRatio can fail on PDF pages or
chrome-extension pages. Using Promise.allSettled ensures the screenshot
still succeeds, falling back to DPR=1.
