diff --git a/packages/browseros-agent/apps/agent/entrypoints/app/agent-command/useAgentConversation.ts b/packages/browseros-agent/apps/agent/entrypoints/app/agent-command/useAgentConversation.ts
index 1f284f12..555f45e4 100644
--- a/packages/browseros-agent/apps/agent/entrypoints/app/agent-command/useAgentConversation.ts
+++ b/packages/browseros-agent/apps/agent/entrypoints/app/agent-command/useAgentConversation.ts
@@ -1,4 +1,4 @@
-import { useEffect, useRef, useState } from 'react'
+import { useCallback, useEffect, useRef, useState } from 'react'
 import {
   type AgentHarnessStreamEvent,
   attachToHarnessTurn,
@@ -16,8 +16,13 @@ import type {
 } from '@/lib/agent-conversations/types'
 import { useInvalidateAgentOutputs } from '@/lib/agent-files'
 import type { ServerAttachmentPayload } from '@/lib/attachments'
+import { sentry } from '@/lib/sentry/sentry'
 import { consumeSSEStream } from '@/lib/sse'
 import { buildToolLabel } from '@/lib/tool-labels'
+import {
+  createWorkflowUsageRecord,
+  recordWorkflowUsage,
+} from '@/lib/workflow-usage/storage'
 import { mapAgentHarnessToolStatus } from './agent-stream-events'
 
 export interface SendInput {
@@ -68,6 +73,8 @@ export function useAgentConversation(
   const streamAbortRef = useRef<AbortController | null>(null)
   const onCompleteRef = useRef(options.onComplete)
   const onSessionKeyChangeRef = useRef(options.onSessionKeyChange)
+  const workflowToolNamesRef = useRef<string[]>([])
+  const workflowToolIdsRef = useRef(new Set<string>())
   // Per-turn resume bookkeeping. `turnId` is captured from the response
   // header; `lastSeq` advances with every SSE event so a reconnect can
   // resume via Last-Event-ID.
@@ -112,6 +119,35 @@ export function useAgentConversation(
     })
   }
 
+  const resetWorkflowUsageCapture = useCallback(() => {
+    workflowToolNamesRef.current = []
+    workflowToolIdsRef.current = new Set()
+  }, [])
+
+  const persistWorkflowUsageCapture = useCallback(
+    (turnId?: string | null) => {
+      const toolNames = workflowToolNamesRef.current
+      if (toolNames.length === 0) return
+
+      void recordWorkflowUsage(
+        createWorkflowUsageRecord({
+          id: `agent-harness-turn:${turnId ?? crypto.randomUUID()}`,
+          source: 'agent-harness-chat',
+          toolNames,
+        }),
+      ).catch((error) => {
+        sentry.captureException(error, {
+          extra: {
+            message: 'Failed to persist agent workflow usage pattern',
+            agentId,
+            turnId,
+          },
+        })
+      })
+    },
+    [agentId],
+  )
+
   const appendTextDelta = (delta: string) => {
     textAccRef.current += delta
     const text = textAccRef.current
@@ -174,6 +210,11 @@ export function useAgentConversation(
   const upsertAgentHarnessTool = (event: AgentHarnessStreamEvent) => {
     if (event.type !== 'tool_call') return
     const rawName = event.title || event.rawType || 'tool call'
+    const toolId = event.id ?? rawName
+    if (!workflowToolIdsRef.current.has(toolId)) {
+      workflowToolIdsRef.current.add(toolId)
+      workflowToolNamesRef.current.push(rawName)
+    }
     const { label, subject } = buildToolLabel(
       rawName,
       event.text ? { description: event.text } : undefined,
@@ -295,6 +336,7 @@ export function useAgentConversation(
         streamAbortRef.current = abortController
         setStreaming(true)
         weStartedStream = true
+        resetWorkflowUsageCapture()
 
         const response = await attachToHarnessTurn(agentId, {
           turnId: active.turnId,
@@ -328,6 +370,7 @@ export function useAgentConversation(
         // itself, so resetting here would only cause a brief flicker.
         if (!cancelled && weStartedStream) {
           const finishedTurnId = turnIdRef.current
+          persistWorkflowUsageCapture(finishedTurnId)
           turnIdRef.current = null
           lastSeqRef.current = null
           setStreaming(false)
@@ -344,7 +387,12 @@ export function useAgentConversation(
       cancelled = true
       abortController.abort()
     }
-  }, [agentId, activeTurnIdDep])
+  }, [
+    agentId,
+    activeTurnIdDep,
+    persistWorkflowUsageCapture,
+    resetWorkflowUsageCapture,
+  ])
 
   /**
    * Send the chat request and follow the 409-active-turn redirect
@@ -422,6 +470,7 @@ export function useAgentConversation(
     }
     setTurns((prev) => [...prev, turn])
     setStreaming(true)
+    resetWorkflowUsageCapture()
     textAccRef.current = ''
     thinkAccRef.current = ''
     const abortController = new AbortController()
@@ -466,6 +515,7 @@ export function useAgentConversation(
       // useAgentTurnFiles consumers also flush, not just the agent-wide
       // rail query.
       const finishedTurnId = turnIdRef.current
+      persistWorkflowUsageCapture(finishedTurnId)
       turnIdRef.current = null
       lastSeqRef.current = null
       onCompleteRef.current?.()
diff --git a/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useChatSession.ts b/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useChatSession.ts
index 11eded17..9849c7fd 100644
--- a/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useChatSession.ts
+++ b/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useChatSession.ts
@@ -48,6 +48,18 @@ import {
   normalizeToolApprovalConfig,
   toolApprovalConfigStorage,
 } from '@/lib/tool-approvals/storage'
+import {
+  analyzeWorkflowUsage,
+  detectWorkflowAdvisorCommand,
+  formatWorkflowAnalysisResponse,
+  formatWorkflowUsageClearedResponse,
+  formatWorkflowUsageDataResponse,
+  type WorkflowAdvisorCommand,
+} from '@/lib/workflow-usage/advisor'
+import {
+  clearWorkflowUsageRecords,
+  getWorkflowUsageRecords,
+} from '@/lib/workflow-usage/storage'
 import { selectedWorkspaceStorage } from '@/lib/workspace/workspace-storage'
 import type { ChatMode } from './chatTypes'
 import { GetConversationWithMessagesDocument } from './graphql/chatSessionDocument'
@@ -133,6 +145,7 @@ export interface ChatSessionOptions {
 }
 
 const NEWTAB_SYSTEM_PROMPT = `IMPORTANT: The user is chatting from the New Tab page. When performing browser actions, ALWAYS open content in a NEW TAB rather than navigating the current tab. The user's new tab page should remain accessible.`
+const WORKFLOW_ADVISOR_LOCAL_ONLY = 'workflow-advisor'
 
 const getUserSystemPrompt = (
   origin: ChatOrigin | undefined,
@@ -142,6 +155,25 @@ const getUserSystemPrompt = (
     ? [personalization, NEWTAB_SYSTEM_PROMPT].filter(Boolean).join('\n\n')
     : personalization
 
+const createTextMessage = (
+  role: 'user' | 'assistant',
+  text: string,
+  options?: { localOnly?: boolean },
+): UIMessage =>
+  ({
+    id: crypto.randomUUID(),
+    role,
+    parts: [{ type: 'text', text }],
+    metadata: options?.localOnly
+      ? { browserosLocalOnly: WORKFLOW_ADVISOR_LOCAL_ONLY }
+      : undefined,
+  }) as UIMessage
+
+const isWorkflowAdvisorLocalOnlyMessage = (message: UIMessage): boolean => {
+  const metadata = (message as { metadata?: Record<string, unknown> }).metadata
+  return metadata?.browserosLocalOnly === WORKFLOW_ADVISOR_LOCAL_ONLY
+}
+
 const buildRequestBrowserContext = ({
   activeTab,
   action,
@@ -376,7 +408,9 @@ export const useChatSession = (options?: ChatSessionOptions) => {
           Feature.PREVIOUS_CONVERSATION_ARRAY,
         )
 
-        const previousMessages = messagesRef.current
+        const previousMessages = messagesRef.current.filter(
+          (message) => !isWorkflowAdvisorLocalOnlyMessage(message),
+        )
         const history =
           previousMessages.length > 0
             ? formatConversationHistory(previousMessages)
@@ -559,7 +593,9 @@ export const useChatSession = (options?: ChatSessionOptions) => {
       })
     }
 
-    const messagesToSave = messages.filter((m) => m.parts?.length > 0)
+    const messagesToSave = messages.filter(
+      (m) => m.parts?.length > 0 && !isWorkflowAdvisorLocalOnlyMessage(m),
+    )
     if (messagesToSave.length === 0) return
 
     if (isLoggedIn) {
@@ -645,6 +681,54 @@ export const useChatSession = (options?: ChatSessionOptions) => {
     action?: ChatAction
   } | null>(null)
 
+  const appendLocalWorkflowAdvisorExchange = (
+    userText: string,
+    responseText: string,
+  ) => {
+    const nextMessages = [
+      ...messagesRef.current,
+      createTextMessage('user', userText, { localOnly: true }),
+      createTextMessage('assistant', responseText, { localOnly: true }),
+    ]
+    messagesRef.current = nextMessages
+    setMessages(nextMessages)
+  }
+
+  const handleWorkflowAdvisorCommand = async (
+    text: string,
+    command: WorkflowAdvisorCommand,
+  ) => {
+    try {
+      if (command === 'clear') {
+        await clearWorkflowUsageRecords()
+        appendLocalWorkflowAdvisorExchange(
+          text,
+          formatWorkflowUsageClearedResponse(),
+        )
+        return
+      }
+
+      const records = await getWorkflowUsageRecords()
+      const response =
+        command === 'view'
+          ? formatWorkflowUsageDataResponse(records)
+          : formatWorkflowAnalysisResponse(analyzeWorkflowUsage(records))
+
+      appendLocalWorkflowAdvisorExchange(text, response)
+    } catch (error) {
+      sentry.captureException(error, {
+        extra: {
+          message: 'Failed to run local workflow advisor command',
+          command,
+        },
+      })
+      appendLocalWorkflowAdvisorExchange(
+        text,
+        "I couldn't read the local workflow usage patterns. Nothing was sent to a model or external service.",
+      )
+    }
+  }
+
   const dispatchMessage = useCallback(
     (text: string) => {
       startExecutionTask({
@@ -696,6 +780,12 @@ export const useChatSession = (options?: ChatSessionOptions) => {
         selectedLlmProvider?.modelId,
     })
 
+    const workflowAdvisorCommand = detectWorkflowAdvisorCommand(params.text)
+    if (workflowAdvisorCommand) {
+      void handleWorkflowAdvisorCommand(params.text, workflowAdvisorCommand)
+      return
+    }
+
     if (!isIntegrationsSyncedRef.current) {
       // Queue the message — will be sent when sync completes
       pendingMessageRef.current = params
diff --git a/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useExecutionHistoryTracker.ts b/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useExecutionHistoryTracker.ts
index b47015b2..0246f532 100644
--- a/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useExecutionHistoryTracker.ts
+++ b/packages/browseros-agent/apps/agent/entrypoints/sidepanel/index/useExecutionHistoryTracker.ts
@@ -10,6 +10,10 @@ import type {
   ExecutionTaskStatus,
 } from '@/lib/execution-history/types'
 import { sentry } from '@/lib/sentry/sentry'
+import {
+  createWorkflowUsageRecordFromExecutionTask,
+  recordWorkflowUsage,
+} from '@/lib/workflow-usage/storage'
 
 interface StartExecutionTaskInput {
   conversationId: string
@@ -145,6 +149,17 @@ export function useExecutionHistoryTracker() {
       }
 
       persistTask(nextTask)
+      void recordWorkflowUsage(
+        createWorkflowUsageRecordFromExecutionTask(nextTask),
+      ).catch((error) => {
+        sentry.captureException(error, {
+          extra: {
+            message: 'Failed to persist workflow usage pattern',
+            conversationId: nextTask.conversationId,
+            taskId: nextTask.id,
+          },
+        })
+      })
       activeTaskRef.current = null
     },
     [persistTask],
diff --git a/packages/browseros-agent/apps/agent/lib/workflow-usage/advisor.test.ts b/packages/browseros-agent/apps/agent/lib/workflow-usage/advisor.test.ts
new file mode 100644
index 00000000..0332384c
--- /dev/null
+++ b/packages/browseros-agent/apps/agent/lib/workflow-usage/advisor.test.ts
@@ -0,0 +1,73 @@
+import { describe, expect, it } from 'bun:test'
+import {
+  analyzeWorkflowUsage,
+  detectWorkflowAdvisorCommand,
+  formatWorkflowAnalysisResponse,
+  normalizeToolSequence,
+} from './advisor'
+import type { WorkflowUsageRecord } from './types'
+
+describe('workflow usage advisor', () => {
+  it('detects explicit workflow advisor commands only', () => {
+    expect(detectWorkflowAdvisorCommand('analyze my workflow')).toBe('analyze')
+    expect(detectWorkflowAdvisorCommand('what patterns do you see?')).toBe(
+      'analyze',
+    )
+    expect(detectWorkflowAdvisorCommand('show workflow usage data')).toBe(
+      'view',
+    )
+    expect(detectWorkflowAdvisorCommand('clear skill suggestion data')).toBe(
+      'clear',
+    )
+    expect(detectWorkflowAdvisorCommand('summarize this page')).toBeNull()
+  })
+
+  it('normalizes command sequences without retaining repeated adjacent tools', () => {
+    expect(normalizeToolSequence([' new_page ', 'new_page', 'open'])).toEqual([
+      'new_page',
+      'open',
+    ])
+  })
+
+  it('suggests repeated local tool-name patterns', () => {
+    const analysis = analyzeWorkflowUsage([
+      record('1', ['new_page', 'navigate', 'get_page_content'], 100),
+      record('2', ['new_page', 'navigate', 'get_page_content'], 200),
+      record('3', ['search', 'open'], 300),
+    ])
+
+    expect(analysis.totalRuns).toBe(3)
+    expect(analysis.suggestions).toHaveLength(1)
+    expect(analysis.suggestions[0]).toMatchObject({
+      runCount: 2,
+      pattern: ['new_page', 'navigate', 'get_page_content'],
+    })
+  })
+
+  it('formats concrete suggestions with a privacy note', () => {
+    const response = formatWorkflowAnalysisResponse(
+      analyzeWorkflowUsage([
+        record('1', ['new_page', 'navigate', 'get_page_content'], 100),
+        record('2', ['new_page', 'navigate', 'get_page_content'], 200),
+      ]),
+    )
+
+    expect(response).toContain('Pattern: Open page -> Navigate -> Read page')
+    expect(response).toContain('Create a "Open Page to Read Page Skill" skill')
+    expect(response).toContain('does not include URLs')
+    expect(response).toContain('tool inputs')
+  })
+})
+
+function record(
+  id: string,
+  toolNames: string[],
+  recordedAt: number,
+): WorkflowUsageRecord {
+  return {
+    id,
+    source: 'sidepanel-chat',
+    recordedAt,
+    toolNames,
+  }
+}
diff --git a/packages/browseros-agent/apps/agent/lib/workflow-usage/advisor.ts b/packages/browseros-agent/apps/agent/lib/workflow-usage/advisor.ts
new file mode 100644
index 00000000..9c5579d6
--- /dev/null
+++ b/packages/browseros-agent/apps/agent/lib/workflow-usage/advisor.ts
@@ -0,0 +1,241 @@
+import type {
+  WorkflowSkillSuggestion,
+  WorkflowUsageAnalysis,
+  WorkflowUsageRecord,
+} from './types'
+
+export type WorkflowAdvisorCommand = 'analyze' | 'view' | 'clear'
+
+const DEFAULT_MIN_RUNS = 2
+const DEFAULT_SUGGESTION_LIMIT = 3
+const MIN_PATTERN_LENGTH = 2
+const MAX_PATTERN_LENGTH = 8
+
+const PRIVACY_NOTE =
+  'This analysis uses only local tool-name sequences. BrowserOS does not include URLs, page content, prompts, tool inputs, or tool outputs in this workflow pattern data.'
+
+const TOOL_LABELS: Record<string, string> = {
+  click: 'Click',
+  extract_data: 'Extract data',
+  filesystem_read: 'Read file',
+  filesystem_write: 'Write file',
+  get_page_content: 'Read page',
+  new_page: 'Open page',
+  navigate: 'Navigate',
+  open: 'Open',
+  screenshot: 'Screenshot',
+  search: 'Search',
+  type: 'Type',
+}
+
+function normalizeCommandText(text: string): string {
+  return text.toLowerCase().replace(/\s+/g, ' ').trim()
+}
+
+export function detectWorkflowAdvisorCommand(
+  text: string,
+): WorkflowAdvisorCommand | null {
+  const normalized = normalizeCommandText(text)
+  if (!normalized) return null
+
+  const mentionsWorkflowData =
+    normalized.includes('workflow usage') ||
+    normalized.includes('usage pattern') ||
+    normalized.includes('workflow pattern') ||
+    normalized.includes('skill suggestion')
+
+  if (
+    mentionsWorkflowData &&
+    /\b(clear|delete|reset|forget)\b/.test(normalized)
+  ) {
+    return 'clear'
+  }
+
+  if (
+    mentionsWorkflowData &&
+    /\b(show|view|list|display|what)\b/.test(normalized)
+  ) {
+    return 'view'
+  }
+
+  if (
+    normalized.includes('analyze my workflow') ||
+    normalized.includes('analyse my workflow') ||
+    normalized.includes('what patterns do you see') ||
+    normalized.includes('suggest skills') ||
+    normalized.includes('find skill suggestions') ||
+    normalized.includes('what can be automated') ||
+    normalized.includes('analyze workflow patterns') ||
+    normalized.includes('analyse workflow patterns')
+  ) {
+    return 'analyze'
+  }
+
+  return null
+}
+
+function normalizeToolName(toolName: string): string {
+  return toolName.trim().toLowerCase()
+}
+
+export function normalizeToolSequence(toolNames: string[]): string[] {
+  const sequence: string[] = []
+  for (const rawName of toolNames) {
+    const toolName = normalizeToolName(rawName)
+    if (!toolName) continue
+    if (sequence[sequence.length - 1] === toolName) continue
+    sequence.push(toolName)
+  }
+  return sequence.slice(0, MAX_PATTERN_LENGTH)
+}
+
+function labelTool(toolName: string): string {
+  const normalized = normalizeToolName(toolName)
+  const known = TOOL_LABELS[normalized]
+  if (known) return known
+
+  return normalized
+    .replace(/^(browser|tool|mcp)[_-]/, '')
+    .replace(/[_-]+/g, ' ')
+    .replace(/\b\w/g, (char) => char.toUpperCase())
+}
+
+function titleCase(value: string): string {
+  return value.replace(/\b\w/g, (char) => char.toUpperCase())
+}
+
+function buildSuggestionTitle(pattern: string[]): string {
+  const labels = pattern.map((toolName) => titleCase(labelTool(toolName)))
+  const uniqueLabels = Array.from(new Set(labels))
+  if (uniqueLabels.length <= 2) return `${uniqueLabels.join(' + ')} Skill`
+  return `${uniqueLabels[0]} to ${uniqueLabels.at(-1)} Skill`
+}
+
+function buildBenefit(pattern: string[]): string {
+  const patternLabel = pattern.map(labelTool).join(' -> ')
+  return `Turn the repeated ${patternLabel} sequence into reusable skill instructions.`
+}
+
+function compareSuggestions(
+  left: WorkflowSkillSuggestion,
+  right: WorkflowSkillSuggestion,
+): number {
+  return (
+    right.runCount - left.runCount ||
+    right.pattern.length - left.pattern.length ||
+    right.lastUsedAt - left.lastUsedAt
+  )
+}
+
+export function analyzeWorkflowUsage(
+  records: WorkflowUsageRecord[],
+  options?: { minRuns?: number; limit?: number },
+): WorkflowUsageAnalysis {
+  const minRuns = options?.minRuns ?? DEFAULT_MIN_RUNS
+  const limit = options?.limit ?? DEFAULT_SUGGESTION_LIMIT
+  const groups = new Map<
+    string,
+    { pattern: string[]; runCount: number; lastUsedAt: number }
+  >()
+
+  for (const record of records) {
+    const pattern = normalizeToolSequence(record.toolNames)
+    if (pattern.length < MIN_PATTERN_LENGTH) continue
+
+    const key = pattern.join('\u001f')
+    const existing = groups.get(key)
+    groups.set(key, {
+      pattern,
+      runCount: (existing?.runCount ?? 0) + 1,
+      lastUsedAt: Math.max(existing?.lastUsedAt ?? 0, record.recordedAt),
+    })
+  }
+
+  const suggestions = Array.from(groups.values())
+    .filter((group) => group.runCount >= minRuns)
+    .map((group, index): WorkflowSkillSuggestion => {
+      const pattern = group.pattern
+      return {
+        id: `workflow-${index + 1}`,
+        title: buildSuggestionTitle(pattern),
+        runCount: group.runCount,
+        pattern,
+        lastUsedAt: group.lastUsedAt,
+        benefit: buildBenefit(pattern),
+      }
+    })
+    .sort(compareSuggestions)
+    .slice(0, limit)
+
+  return {
+    totalRuns: records.length,
+    eligibleRuns: Array.from(groups.values()).reduce(
+      (count, group) => count + group.runCount,
+      0,
+    ),
+    suggestions,
+  }
+}
+
+export function formatWorkflowAnalysisResponse(
+  analysis: WorkflowUsageAnalysis,
+): string {
+  if (analysis.suggestions.length === 0) {
+    return [
+      "I don't have enough repeated local tool patterns to suggest a custom skill yet.",
+      '',
+      PRIVACY_NOTE,
+      '',
+      `Tracked runs: ${analysis.totalRuns}. Eligible repeated-tool runs: ${analysis.eligibleRuns}.`,
+    ].join('\n')
+  }
+
+  const suggestionLines = analysis.suggestions.map((suggestion, index) =>
+    [
+      `${index + 1}. ${suggestion.title} -> ${suggestion.runCount} times`,
+      `   Pattern: ${suggestion.pattern.map(labelTool).join(' -> ')}`,
+      `   Suggestion: Create a "${suggestion.title}" skill for this exact command sequence.`,
+      `   Benefit: ${suggestion.benefit}`,
+    ].join('\n'),
+  )
+
+  return [
+    `Found ${analysis.suggestions.length} potential automation${analysis.suggestions.length === 1 ? '' : 's'}:`,
+    '',
+    ...suggestionLines,
+    '',
+    PRIVACY_NOTE,
+  ].join('\n')
+}
+
+export function formatWorkflowUsageDataResponse(
+  records: WorkflowUsageRecord[],
+): string {
+  if (records.length === 0) {
+    return [
+      'No local workflow usage patterns are stored yet.',
+      '',
+      PRIVACY_NOTE,
+    ].join('\n')
+  }
+
+  const recentRecords = records
+    .slice()
+    .sort((left, right) => right.recordedAt - left.recordedAt)
+    .slice(0, 10)
+
+  return [
+    `Stored local workflow runs: ${records.length}`,
+    '',
+    ...recentRecords.map(
+      (record, index) =>
+        `${index + 1}. ${normalizeToolSequence(record.toolNames).map(labelTool).join(' -> ')}`,
+    ),
+    '',
+    PRIVACY_NOTE,
+  ].join('\n')
+}
+
+export function formatWorkflowUsageClearedResponse(): string {
+  return 'Cleared the local workflow usage pattern data. No URLs, page content, prompts, tool inputs, or tool outputs were stored in this data.'
+}
diff --git a/packages/browseros-agent/apps/agent/lib/workflow-usage/storage.ts b/packages/browseros-agent/apps/agent/lib/workflow-usage/storage.ts
new file mode 100644
index 00000000..a2ad4407
--- /dev/null
+++ b/packages/browseros-agent/apps/agent/lib/workflow-usage/storage.ts
@@ -0,0 +1,78 @@
+import { storage } from '@wxt-dev/storage'
+import type { ExecutionTaskRecord } from '@/lib/execution-history/types'
+import { normalizeToolSequence } from './advisor'
+import type {
+  WorkflowUsageRecord,
+  WorkflowUsageSource,
+  WorkflowUsageStore,
+} from './types'
+
+const MAX_WORKFLOW_USAGE_RECORDS = 300
+
+export const workflowUsageStorage = storage.defineItem<WorkflowUsageStore>(
+  'local:workflowUsagePatterns',
+  {
+    fallback: { version: 1, records: [] },
+    version: 1,
+  },
+)
+
+export function createWorkflowUsageRecord(input: {
+  id: string
+  source: WorkflowUsageSource
+  toolNames: string[]
+  recordedAt?: number
+}): WorkflowUsageRecord | null {
+  const toolNames = normalizeToolSequence(input.toolNames)
+  if (toolNames.length === 0) return null
+
+  return {
+    id: input.id,
+    source: input.source,
+    recordedAt: input.recordedAt ?? Date.now(),
+    toolNames,
+  }
+}
+
+export function createWorkflowUsageRecordFromExecutionTask(
+  task: ExecutionTaskRecord,
+): WorkflowUsageRecord | null {
+  return createWorkflowUsageRecord({
+    id: `execution-task:${task.id}`,
+    source: 'sidepanel-chat',
+    recordedAt: Date.parse(task.completedAt ?? task.startedAt),
+    toolNames: task.steps.map((step) => step.toolName),
+  })
+}
+
+export async function recordWorkflowUsage(
+  record: WorkflowUsageRecord | null,
+): Promise<void> {
+  if (!record) return
+
+  const current = (await workflowUsageStorage.getValue()) ?? {
+    version: 1,
+    records: [],
+  }
+  const recordsById = new Map(
+    current.records.map((existing) => [existing.id, existing]),
+  )
+  recordsById.set(record.id, record)
+
+  const records = Array.from(recordsById.values())
+    .sort((left, right) => left.recordedAt - right.recordedAt)
+    .slice(-MAX_WORKFLOW_USAGE_RECORDS)
+
+  await workflowUsageStorage.setValue({ version: 1, records })
+}
+
+export async function getWorkflowUsageRecords(): Promise<
+  WorkflowUsageRecord[]
+> {
+  const current = await workflowUsageStorage.getValue()
+  return current?.records ?? []
+}
+
+export async function clearWorkflowUsageRecords(): Promise<void> {
+  await workflowUsageStorage.setValue({ version: 1, records: [] })
+}
diff --git a/packages/browseros-agent/apps/agent/lib/workflow-usage/types.ts b/packages/browseros-agent/apps/agent/lib/workflow-usage/types.ts
new file mode 100644
index 00000000..a2098f7f
--- /dev/null
+++ b/packages/browseros-agent/apps/agent/lib/workflow-usage/types.ts
@@ -0,0 +1,28 @@
+export type WorkflowUsageSource = 'sidepanel-chat' | 'agent-harness-chat'
+
+export interface WorkflowUsageRecord {
+  id: string
+  source: WorkflowUsageSource
+  recordedAt: number
+  toolNames: string[]
+}
+
+export interface WorkflowUsageStore {
+  version: 1
+  records: WorkflowUsageRecord[]
+}
+
+export interface WorkflowSkillSuggestion {
+  id: string
+  title: string
+  runCount: number
+  pattern: string[]
+  lastUsedAt: number
+  benefit: string
+}
+
+export interface WorkflowUsageAnalysis {
+  totalRuns: number
+  eligibleRuns: number
+  suggestions: WorkflowSkillSuggestion[]
+}