fix(ai-vault): surface a workspace's own sessions past the recency cap (#6273)

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Orca <help@stably.ai>
Co-authored-by: brennanb2025 <brennankbenson@gmail.com>

src/main/ai-vault/session-scanner-scope-discovery.ts

@@ -0,0 +1,253 @@
+import { createReadStream } from 'fs'
+import { readdir, stat } from 'fs/promises'
+import { createInterface } from 'readline'
+import { extname, join } from 'path'
+import {
+  isPathInsideOrEqual,
+  normalizeRuntimePathForComparison
+} from '../../shared/cross-platform-path'
+import type { AiVaultScanIssue } from '../../shared/ai-vault-types'
+import { parseWslUncPath } from '../../shared/wsl-paths'
+import type { FileWithMtime } from './session-scanner-types'
+import { errorMessage, extractString, parseJsonObject } from './session-scanner-values'
+
+// Reading a few lines of one transcript per project dir is enough to learn that
+// dir's cwd; cap both so a giant or cwd-less transcript can't stall the scan.
+const REPRESENTATIVE_CWD_LINE_LIMIT = 200
+const REPRESENTATIVE_FILE_LIMIT = 3
+const CLAUDE_EXTENSIONS = new Set(['.jsonl'])
+
+/**
+ * Fully include the transcripts of Claude project directories whose cwd falls
+ * inside the active workspace/project paths.
+ *
+ * Why: Claude organizes `~/.claude/projects/<cwd-encoded>/` one directory per
+ * cwd. The global scan is recency-capped, so a project the user hasn't touched
+ * recently can drop off the list entirely even though `claude --resume` still
+ * finds it. For scoped panel views we resolve each project dir's cwd cheaply and
+ * bypass the cap for the ones that belong to the active scope.
+ */
+export async function discoverInScopeClaudeFiles(args: {
+  rootDirs: readonly string[]
+  scopePaths: readonly string[]
+  limit: number
+  excludedFilePaths: ReadonlySet<string>
+  issues: AiVaultScanIssue[]
+}): Promise<FileWithMtime[]> {
+  if (args.scopePaths.length === 0 || args.limit <= 0) {
+    return []
+  }
+  const scopeProjectPrefixes = claudeProjectScopePrefixes(args.scopePaths)
+  const collected = new Map<string, FileWithMtime>()
+  for (const rootDir of args.rootDirs) {
+    for (const projectDir of await listProjectDirs(rootDir, scopeProjectPrefixes)) {
+      const cwd = await readProjectDirCwd(projectDir)
+      if (!cwd || !args.scopePaths.some((scopePath) => isCwdInsideScopePath(scopePath, cwd))) {
+        continue
+      }
+      await collectClaudeFiles({
+        projectDir,
+        issues: args.issues,
+        collected,
+        limit: args.limit,
+        excludedFilePaths: args.excludedFilePaths
+      })
+    }
+  }
+  return [...collected.values()].sort((left, right) => right.mtimeMs - left.mtimeMs)
+}
+
+function claudeProjectScopePrefixes(scopePaths: readonly string[]): Set<string> {
+  const prefixes = new Set<string>()
+  for (const scopePath of scopePaths) {
+    for (const candidate of scopePathCandidates(scopePath)) {
+      prefixes.add(encodeClaudeProjectPath(candidate))
+    }
+  }
+  return prefixes
+}
+
+function scopePathCandidates(scopePath: string): string[] {
+  const wslScopePath = parseWslUncPath(scopePath)
+  return wslScopePath ? [scopePath, wslScopePath.linuxPath] : [scopePath]
+}
+
+function encodeClaudeProjectPath(pathValue: string): string {
+  return normalizeRuntimePathForComparison(pathValue).replace(/[^a-zA-Z0-9]/g, '-')
+}
+
+function isClaudeProjectDirInScope(projectDirName: string, scopePrefixes: ReadonlySet<string>) {
+  for (const prefix of scopePrefixes) {
+    if (projectDirName === prefix || projectDirName.startsWith(`${prefix}-`)) {
+      return true
+    }
+  }
+  return false
+}
+
+function isCwdInsideScopePath(scopePath: string, cwd: string): boolean {
+  if (isPathInsideOrEqual(scopePath, cwd)) {
+    return true
+  }
+
+  const wslScopePath = parseWslUncPath(scopePath)
+  if (!wslScopePath) {
+    return false
+  }
+
+  // WSL transcripts record Linux cwd values even when the renderer sends the
+  // active worktree as a Windows UNC path.
+  return isPathInsideOrEqual(wslScopePath.linuxPath, cwd)
+}
+
+async function listProjectDirs(
+  rootDir: string,
+  scopeProjectPrefixes: ReadonlySet<string>
+): Promise<string[]> {
+  let entries
+  try {
+    entries = await readdir(rootDir, { withFileTypes: true })
+  } catch {
+    return []
+  }
+  return entries
+    .filter(
+      (entry) => entry.isDirectory() && isClaudeProjectDirInScope(entry.name, scopeProjectPrefixes)
+    )
+    .map((entry) => join(rootDir, entry.name))
+}
+
+async function readProjectDirCwd(projectDir: string): Promise<string | null> {
+  const files = await newestClaudeFilesInDir(projectDir)
+  for (const file of files.slice(0, REPRESENTATIVE_FILE_LIMIT)) {
+    const cwd = await readFirstCwd(file)
+    if (cwd) {
+      return cwd
+    }
+  }
+  return null
+}
+
+async function newestClaudeFilesInDir(projectDir: string): Promise<string[]> {
+  let entries
+  try {
+    entries = await readdir(projectDir, { withFileTypes: true })
+  } catch {
+    return []
+  }
+  const newest: { path: string; mtimeMs: number }[] = []
+  for (const entry of entries) {
+    if (!entry.isFile() || !CLAUDE_EXTENSIONS.has(extname(entry.name).toLowerCase())) {
+      continue
+    }
+    const path = join(projectDir, entry.name)
+    try {
+      addBoundedPath(newest, REPRESENTATIVE_FILE_LIMIT, {
+        path,
+        mtimeMs: (await stat(path)).mtimeMs
+      })
+    } catch {
+      // Best effort: unreadable candidates are ignored here and reported during
+      // full collection if the project directory proves in-scope.
+    }
+  }
+  return newest.sort((left, right) => right.mtimeMs - left.mtimeMs).map((value) => value.path)
+}
+
+async function readFirstCwd(filePath: string): Promise<string | null> {
+  const input = createReadStream(filePath, { encoding: 'utf-8' })
+  const lines = createInterface({ input, crlfDelay: Infinity })
+  let read = 0
+  try {
+    for await (const line of lines) {
+      if (read++ >= REPRESENTATIVE_CWD_LINE_LIMIT) {
+        break
+      }
+      const cwd = extractString(parseJsonObject(line)?.cwd)
+      if (cwd) {
+        return cwd
+      }
+    }
+  } catch {
+    return null
+  } finally {
+    // readline.close() leaves the underlying stream open; destroy it so the early
+    // break/catch paths don't leak a file descriptor (this runs per project dir).
+    lines.close()
+    input.destroy()
+  }
+  return null
+}
+
+async function collectClaudeFiles(args: {
+  projectDir: string
+  issues: AiVaultScanIssue[]
+  collected: Map<string, FileWithMtime>
+  limit: number
+  excludedFilePaths: ReadonlySet<string>
+}): Promise<void> {
+  let entries
+  try {
+    entries = await readdir(args.projectDir, { withFileTypes: true })
+  } catch {
+    return
+  }
+  for (const entry of entries) {
+    if (!entry.isFile() || !CLAUDE_EXTENSIONS.has(extname(entry.name).toLowerCase())) {
+      continue
+    }
+    const path = join(args.projectDir, entry.name)
+    if (args.collected.has(path) || args.excludedFilePaths.has(path)) {
+      continue
+    }
+    try {
+      const fileStat = await stat(path)
+      addBoundedFile(args.collected, args.limit, {
+        path,
+        mtimeMs: fileStat.mtimeMs,
+        modifiedAt: fileStat.mtime.toISOString()
+      })
+    } catch (err) {
+      args.issues.push({ agent: 'claude', path, message: errorMessage(err) })
+    }
+  }
+}
+
+function addBoundedFile(
+  collected: Map<string, FileWithMtime>,
+  limit: number,
+  file: FileWithMtime
+): void {
+  if (collected.size < limit) {
+    collected.set(file.path, file)
+    return
+  }
+
+  let oldest: FileWithMtime | null = null
+  for (const candidate of collected.values()) {
+    if (!oldest || candidate.mtimeMs < oldest.mtimeMs) {
+      oldest = candidate
+    }
+  }
+  if (oldest && file.mtimeMs > oldest.mtimeMs) {
+    collected.delete(oldest.path)
+    collected.set(file.path, file)
+  }
+}
+
+function addBoundedPath<T extends { mtimeMs: number }>(items: T[], limit: number, item: T): void {
+  if (items.length < limit) {
+    items.push(item)
+    return
+  }
+
+  let oldestIndex = 0
+  for (let index = 1; index < items.length; index++) {
+    if (items[index].mtimeMs < items[oldestIndex].mtimeMs) {
+      oldestIndex = index
+    }
+  }
+  if (item.mtimeMs > items[oldestIndex].mtimeMs) {
+    items[oldestIndex] = item
+  }
+}

src/main/ai-vault/session-scanner-scope.test.ts

@@ -0,0 +1,169 @@
+import { mkdtemp, mkdir, rm, utimes, writeFile } from 'fs/promises'
+import { tmpdir } from 'os'
+import { join } from 'path'
+import { afterEach, describe, expect, it } from 'vitest'
+import { scanAiVaultSessions } from './session-scanner'
+import type { AiVaultScanOptions } from './session-scanner-types'
+
+let tempRoots: string[] = []
+
+afterEach(async () => {
+  await Promise.all(tempRoots.map((root) => rm(root, { recursive: true, force: true })))
+  tempRoots = []
+})
+
+// Point every non-Claude source at a nonexistent dir so the scan only sees the
+// Claude fixtures created per test.
+function scopedScanOptions(claudeProjectsDir: string, extra: Partial<AiVaultScanOptions>) {
+  return {
+    claudeProjectsDir,
+    codexSessionsDir: '/nonexistent/codex',
+    geminiSessionsDir: '/nonexistent/gemini',
+    copilotSessionsDir: '/nonexistent/copilot',
+    cursorProjectsDir: '/nonexistent/cursor',
+    opencodeStorageDir: '/nonexistent/opencode',
+    opencodeDbPaths: [],
+    grokSessionsDir: '/nonexistent/grok',
+    devinTranscriptsDir: '/nonexistent/devin',
+    hermesSessionsDir: '/nonexistent/hermes',
+    rovoSessionsDir: '/nonexistent/rovo',
+    openclawStateDir: '/nonexistent/openclaw',
+    openclawLegacyStateDir: '/nonexistent/openclaw-legacy',
+    piSessionsDir: '/nonexistent/pi',
+    droidSessionsDir: '/nonexistent/droid',
+    droidProjectsDir: '/nonexistent/droid-projects',
+    kimiSessionsDir: '/nonexistent/kimi',
+    ...extra
+  } satisfies AiVaultScanOptions
+}
+
+async function writeClaudeSession(args: {
+  claudeRoot: string
+  dirName: string
+  sessionId: string
+  cwd: string
+  iso: string
+  leadingCwdlessLine?: boolean
+}): Promise<void> {
+  const dir = join(args.claudeRoot, args.dirName)
+  await mkdir(dir, { recursive: true })
+  const records: unknown[] = []
+  if (args.leadingCwdlessLine) {
+    records.push({ type: 'last-prompt', sessionId: args.sessionId })
+  }
+  records.push({
+    type: 'user',
+    sessionId: args.sessionId,
+    timestamp: args.iso,
+    cwd: args.cwd,
+    message: { role: 'user', content: `work in ${args.cwd}` }
+  })
+  const filePath = join(dir, `${args.sessionId}.jsonl`)
+  await writeFile(filePath, records.map((record) => JSON.stringify(record)).join('\n'))
+  const time = new Date(args.iso)
+  await utimes(filePath, time, time)
+}
+
+describe('scanAiVaultSessions scope inclusion', () => {
+  it('surfaces in-scope sessions older than the global recency cap', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'orca-ai-vault-scope-'))
+    tempRoots.push(root)
+    const claudeRoot = join(root, 'claude-projects')
+
+    // One old in-scope session that the recency cap would otherwise drop.
+    await writeClaudeSession({
+      claudeRoot,
+      dirName: '-repo-app',
+      sessionId: 'old-in-scope',
+      cwd: '/repo/app',
+      iso: '2026-01-01T00:00:00.000Z',
+      leadingCwdlessLine: true
+    })
+    // A session in a sub-cwd directory of the same workspace path.
+    await writeClaudeSession({
+      claudeRoot,
+      dirName: '-repo-app-packages-ui',
+      sessionId: 'old-in-scope-subdir',
+      cwd: '/repo/app/packages/ui',
+      iso: '2026-01-02T00:00:00.000Z'
+    })
+    // Recent out-of-scope sessions that fill the cap.
+    for (let index = 0; index < 4; index++) {
+      await writeClaudeSession({
+        claudeRoot,
+        dirName: `-other-${index}`,
+        sessionId: `recent-${index}`,
+        cwd: `/other/${index}`,
+        iso: `2026-06-2${index}T00:00:00.000Z`
+      })
+    }
+
+    const withoutScope = await scanAiVaultSessions(scopedScanOptions(claudeRoot, { limit: 2 }))
+    const withScope = await scanAiVaultSessions(
+      scopedScanOptions(claudeRoot, { limit: 2, scopePaths: ['/repo/app'] })
+    )
+
+    const ids = (result: { sessions: { sessionId: string }[] }) =>
+      result.sessions.map((session) => session.sessionId)
+
+    // The cap hides the old in-scope sessions when no scope is provided.
+    expect(ids(withoutScope)).not.toContain('old-in-scope')
+    expect(ids(withoutScope)).not.toContain('old-in-scope-subdir')
+    // Scope paths force them back in, including the sub-cwd directory.
+    expect(ids(withScope)).toContain('old-in-scope')
+    expect(ids(withScope)).toContain('old-in-scope-subdir')
+  })
+
+  it('does not duplicate sessions already in the capped result', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'orca-ai-vault-scope-'))
+    tempRoots.push(root)
+    const claudeRoot = join(root, 'claude-projects')
+
+    await writeClaudeSession({
+      claudeRoot,
+      dirName: '-repo-app',
+      sessionId: 'recent-in-scope',
+      cwd: '/repo/app',
+      iso: '2026-06-24T00:00:00.000Z'
+    })
+
+    const result = await scanAiVaultSessions(
+      scopedScanOptions(claudeRoot, { limit: 50, scopePaths: ['/repo/app'] })
+    )
+
+    const matches = result.sessions.filter((session) => session.sessionId === 'recent-in-scope')
+    expect(matches).toHaveLength(1)
+  })
+
+  it('matches WSL UNC scope paths against Linux Claude cwd values', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'orca-ai-vault-scope-'))
+    tempRoots.push(root)
+    const claudeRoot = join(root, 'claude-projects')
+
+    await writeClaudeSession({
+      claudeRoot,
+      dirName: '-home-ada-repo',
+      sessionId: 'old-wsl-in-scope',
+      cwd: '/home/ada/repo',
+      iso: '2026-01-01T00:00:00.000Z'
+    })
+    for (let index = 0; index < 4; index++) {
+      await writeClaudeSession({
+        claudeRoot,
+        dirName: `-other-wsl-${index}`,
+        sessionId: `recent-wsl-${index}`,
+        cwd: `/other/wsl/${index}`,
+        iso: `2026-06-2${index}T00:00:00.000Z`
+      })
+    }
+
+    const result = await scanAiVaultSessions(
+      scopedScanOptions(claudeRoot, {
+        limit: 2,
+        scopePaths: ['\\\\wsl.localhost\\Ubuntu\\home\\ada\\repo']
+      })
+    )
+
+    expect(result.sessions.map((session) => session.sessionId)).toContain('old-wsl-in-scope')
+  })
+})

src/main/ai-vault/session-scanner-types.ts

@@ -29,6 +29,9 @@ export type AiVaultScanOptions = {
   kimiSessionsDir?: string
   limit?: number
   limitPerAgent?: number
+  // Active workspace/project paths whose sessions must be included regardless of
+  // the recency cap (see discoverInScopeClaudeFiles).
+  scopePaths?: readonly string[]
   platform?: NodeJS.Platform
 }
 

src/main/ai-vault/session-scanner.ts

@@ -6,6 +6,7 @@ import type {
 import { sessionSortTime } from './session-scanner-accumulator'
 import { parseAgentSessionFile } from './session-scanner-agent-parser'
 import { codexHomeForSessionsDir } from './session-scanner-codex-paths'
+import { discoverInScopeClaudeFiles } from './session-scanner-scope-discovery'
 import {
   DEFAULT_CODEX_HOME_DIR,
   discoverAiVaultSessionSources
@@ -13,6 +14,7 @@ import {
 import type {
   AiVaultScanOptions,
   SessionFileCandidate,
+  SessionFileDiscovery,
   SessionParseResult
 } from './session-scanner-types'
 import { clampPositiveInteger, errorMessage } from './session-scanner-values'
@@ -20,6 +22,9 @@ import { clampPositiveInteger, errorMessage } from './session-scanner-values'
 const DEFAULT_LIMIT = 1000
 const DEFAULT_SCAN_LIMIT_PER_AGENT = 1000
 const SESSION_PARSE_CONCURRENCY = 8
+// Upper bound on extra in-scope transcripts discovered and parsed past the
+// recency cap; guards against a pathological scoped history directory.
+const SCOPE_PARSE_LIMIT = 2000
 
 /**
  * Scan all supported AI agent session stores and return a unified, sorted,
@@ -61,17 +66,80 @@ export async function scanAiVaultSessions(
     issues
   })
 
-  const sessions = parsedSessions
+  const cappedSessions = parsedSessions
     .sort((left, right) => sessionSortTime(right) - sessionSortTime(left))
     .slice(0, limit)
 
+  const scopeSessions = await scanInScopeSessions({
+    discoveries,
+    scopePaths: options.scopePaths ?? [],
+    alreadyParsedFilePaths: new Set(cappedSessions.map((session) => session.filePath)),
+    platform,
+    issues
+  })
+
   return {
-    sessions,
+    sessions: mergeSessions(cappedSessions, scopeSessions),
     issues,
     scannedAt: new Date().toISOString()
   }
 }
 
+// In-scope sessions are guaranteed regardless of the recency cap, so the global
+// (already capped) result and the scope result are unioned and de-duplicated by
+// session id, then re-sorted DESC.
+function mergeSessions(
+  cappedSessions: AiVaultSession[],
+  scopeSessions: AiVaultSession[]
+): AiVaultSession[] {
+  if (scopeSessions.length === 0) {
+    return cappedSessions
+  }
+  const byId = new Map<string, AiVaultSession>()
+  for (const session of cappedSessions) {
+    byId.set(session.id, session)
+  }
+  for (const session of scopeSessions) {
+    byId.set(session.id, session)
+  }
+  return [...byId.values()].sort((left, right) => sessionSortTime(right) - sessionSortTime(left))
+}
+
+async function scanInScopeSessions(args: {
+  discoveries: SessionFileDiscovery[]
+  scopePaths: readonly string[]
+  alreadyParsedFilePaths: ReadonlySet<string>
+  platform: NodeJS.Platform
+  issues: AiVaultScanIssue[]
+}): Promise<AiVaultSession[]> {
+  if (args.scopePaths.length === 0) {
+    return []
+  }
+  const claudeRootDirs = args.discoveries
+    .filter((discovery) => discovery.agent === 'claude')
+    .map((discovery) => discovery.rootDir)
+  const files = await discoverInScopeClaudeFiles({
+    rootDirs: claudeRootDirs,
+    scopePaths: args.scopePaths,
+    limit: SCOPE_PARSE_LIMIT,
+    excludedFilePaths: args.alreadyParsedFilePaths,
+    issues: args.issues
+  })
+  const candidates = files.map(
+    (file): SessionFileCandidate => ({ agent: 'claude', file, codexHome: null })
+  )
+  if (candidates.length === 0) {
+    return []
+  }
+  // Parse every in-scope candidate (limit === candidate count never early-stops).
+  return parseSessionCandidates({
+    candidates,
+    limit: candidates.length,
+    platform: args.platform,
+    issues: args.issues
+  })
+}
+
 async function parseSessionCandidates(args: {
   candidates: SessionFileCandidate[]
   limit: number

src/main/ipc/ai-vault.ts

@@ -22,7 +22,11 @@ let inflightKey: string | null = null
 let handlerOptions: AiVaultHandlerOptions = {}
 
 async function listAiVaultSessions(args?: AiVaultListArgs): Promise<AiVaultListResult> {
-  const key = String(args?.limit ?? 'default')
+  // Scope paths change the result set, so they must be part of the cache key.
+  const key = JSON.stringify({
+    limit: args?.limit ?? 'default',
+    scopePaths: args?.scopePaths ?? []
+  })
   const now = Date.now()
   // Why: opening this panel repeatedly should not re-parse hundreds of JSONL
   // transcripts; explicit refreshes bypass the cache but not an active scan.
@@ -40,6 +44,7 @@ async function listAiVaultSessions(args?: AiVaultListArgs): Promise<AiVaultListR
   inflightList = (async () =>
     scanAiVaultSessions({
       limit: args?.limit,
+      scopePaths: args?.scopePaths,
       additionalCodexSessionsDirs,
       wslHomeDirs: await getAiVaultWslHomeDirs()
     }))()

src/renderer/src/components/right-sidebar/AiVaultPanel.tsx

@@ -14,12 +14,11 @@ import {
   useRepoById,
   useRepos
 } from '@/store/selectors'
+import { agentLabel, filterAiVaultSessions, groupAiVaultSessions } from './ai-vault-session-filters'
 import {
-  agentLabel,
-  deriveAiVaultWorkspaceScopePaths,
-  filterAiVaultSessions,
-  groupAiVaultSessions
-} from './ai-vault-session-filters'
+  deriveAiVaultScopeSessionPaths,
+  deriveAiVaultWorkspaceScopePaths
+} from './ai-vault-scope-paths'
 import {
   DEFAULT_AI_VAULT_SCOPE,
   getRestorableAiVaultScope,
@@ -30,7 +29,6 @@ import {
   AI_VAULT_AGENTS,
   type AiVaultAgent,
   type AiVaultGroup,
-  type AiVaultListResult,
   type AiVaultScope,
   type AiVaultSession,
   type AiVaultSort
@@ -39,8 +37,7 @@ import { getLocalExecutionHostLabel } from '../../../../shared/execution-host'
 import { translate } from '@/i18n/i18n'
 import { AiVaultPanelHeader } from './AiVaultPanelHeader'
 import { AiVaultSessionVirtualList } from './AiVaultSessionVirtualList'
-
-const SESSION_LIMIT = 500
+import { useAiVaultSessionRefresh } from './ai-vault-session-refresh'
 
 export default function AiVaultPanel(): React.JSX.Element {
   const activeWorktree = useActiveWorktree()
@@ -57,14 +54,7 @@ export default function AiVaultPanel(): React.JSX.Element {
   const [group, setGroup] = useState<AiVaultGroup>('project')
   const [hideEmptySessions, setHideEmptySessions] = useState(true)
   const [agents, setAgents] = useState<AiVaultAgent[]>([...AI_VAULT_AGENTS])
-  const [sessions, setSessions] = useState<AiVaultSession[]>([])
-  const [scanResult, setScanResult] = useState<AiVaultListResult | null>(null)
-  const [loading, setLoading] = useState(false)
-  const [error, setError] = useState<string | null>(null)
   const [collapsedGroups, setCollapsedGroups] = useState<Set<string>>(() => new Set())
-  const refreshIdRef = useRef(0)
-  const refreshInFlightRef = useRef(false)
-  const mountedRef = useRef(true)
   const userChangedScopeRef = useRef(false)
   const preferredScopeRef = useRef<AiVaultScope>(DEFAULT_AI_VAULT_SCOPE)
 
@@ -75,7 +65,31 @@ export default function AiVaultPanel(): React.JSX.Element {
     () => deriveAiVaultWorkspaceScopePaths(activeWorktree ?? null, allWorktrees),
     [activeWorktree, allWorktrees]
   )
-  const projectContext = useMemo(
+  const projectScopeContext = useMemo(
+    () =>
+      buildAiVaultProjectContext({
+        repos,
+        worktrees: allWorktrees,
+        projectHostSetupProjection,
+        activeRepo,
+        activeWorktree,
+        sessions: []
+      }),
+    [activeRepo, activeWorktree, allWorktrees, projectHostSetupProjection, repos]
+  )
+  const activeProjectKey = projectScopeContext.activeProjectKey
+  const projectLabelByKey = projectScopeContext.projectLabelByKey
+  // Sent to the scanner so scoped views surface sessions older than the global cap.
+  const scopePaths = useMemo(
+    () =>
+      deriveAiVaultScopeSessionPaths(activeWorktree ?? null, allWorktrees, {
+        activeProjectKey,
+        projectHostSetupProjection
+      }),
+    [activeProjectKey, activeWorktree, allWorktrees, projectHostSetupProjection]
+  )
+  const { error, loading, refresh, scanResult, sessions } = useAiVaultSessionRefresh(scopePaths)
+  const sessionProjectById = useMemo(
     () =>
       buildAiVaultProjectContext({
         repos,
@@ -84,12 +98,9 @@ export default function AiVaultPanel(): React.JSX.Element {
         activeRepo,
         activeWorktree,
         sessions
-      }),
+      }).sessionProjectById,
     [activeRepo, activeWorktree, allWorktrees, projectHostSetupProjection, repos, sessions]
   )
-  const activeProjectKey = projectContext.activeProjectKey
-  const projectLabelByKey = projectContext.projectLabelByKey
-  const sessionProjectById = projectContext.sessionProjectById
   const hasAllAgentsSelected = agents.length === AI_VAULT_AGENTS.length
   const viewAdjustmentCount =
     (hasAllAgentsSelected ? 0 : 1) +
@@ -122,51 +133,6 @@ export default function AiVaultPanel(): React.JSX.Element {
     }
   }, [activeProjectKey, activeWorktreePath, scope])
 
-  const refresh = useCallback(async (args: { force?: boolean } = {}): Promise<void> => {
-    if (refreshInFlightRef.current) {
-      return
-    }
-
-    refreshInFlightRef.current = true
-    const refreshId = refreshIdRef.current + 1
-    refreshIdRef.current = refreshId
-    setLoading(true)
-    setError(null)
-    try {
-      const result = await window.api.aiVault.listSessions({
-        limit: SESSION_LIMIT,
-        force: args.force
-      })
-      if (!mountedRef.current || refreshIdRef.current !== refreshId) {
-        return
-      }
-      setScanResult(result)
-      setSessions(result.sessions)
-    } catch (err) {
-      if (mountedRef.current && refreshIdRef.current === refreshId) {
-        setError(err instanceof Error ? err.message : String(err))
-      }
-    } finally {
-      refreshInFlightRef.current = false
-      if (mountedRef.current && refreshIdRef.current === refreshId) {
-        setLoading(false)
-      }
-    }
-  }, [])
-
-  useEffect(() => {
-    mountedRef.current = true
-    return () => {
-      mountedRef.current = false
-      refreshIdRef.current += 1
-      refreshInFlightRef.current = false
-    }
-  }, [])
-
-  useEffect(() => {
-    void refresh()
-  }, [refresh])
-
   const filteredSessions = useMemo(
     () =>
       filterAiVaultSessions(sessions, {

src/renderer/src/components/right-sidebar/ai-vault-scope-paths.ts

@@ -0,0 +1,128 @@
+import {
+  isRuntimePathAbsolute,
+  normalizeRuntimePathForComparison
+} from '../../../../shared/cross-platform-path'
+import type { ProjectHostSetupProjection } from '../../../../shared/project-host-setup-projection'
+import type { Worktree } from '../../../../shared/types'
+import { splitWorktreeIdForFilesystem } from '../../../../shared/worktree-id'
+
+export function deriveAiVaultWorkspaceScopePaths(
+  activeWorktree: Pick<Worktree, 'id' | 'path' | 'priorWorktreeIds' | 'repoId'> | null,
+  liveWorktrees: readonly Pick<Worktree, 'id' | 'path' | 'repoId'>[] = []
+): string[] {
+  if (!activeWorktree) {
+    return []
+  }
+
+  const paths: string[] = []
+  addAiVaultWorkspaceScopePath(paths, activeWorktree.path)
+
+  for (const priorWorktreeId of activeWorktree.priorWorktreeIds ?? []) {
+    const parsed = splitWorktreeIdForFilesystem(priorWorktreeId)
+    if (!parsed || parsed.repoId !== activeWorktree.repoId) {
+      continue
+    }
+    if (isAiVaultWorkspaceScopePathClaimed(parsed.worktreePath, activeWorktree, liveWorktrees)) {
+      continue
+    }
+    addAiVaultWorkspaceScopePath(paths, parsed.worktreePath)
+  }
+
+  return paths
+}
+
+/**
+ * Paths sent to the scanner so a scoped panel view surfaces its own sessions
+ * even when they are older than the global recency cap. Covers the active
+ * workspace plus the active project's other worktrees (same repo), so both the
+ * Workspace and Project scopes stay complete.
+ */
+export function deriveAiVaultScopeSessionPaths(
+  activeWorktree: Pick<
+    Worktree,
+    'id' | 'path' | 'priorWorktreeIds' | 'projectId' | 'repoId'
+  > | null,
+  liveWorktrees: readonly Pick<Worktree, 'id' | 'path' | 'projectId' | 'repoId'>[] = [],
+  options: {
+    activeProjectKey?: string | null
+    projectHostSetupProjection?: ProjectHostSetupProjection
+  } = {}
+): string[] {
+  const paths = deriveAiVaultWorkspaceScopePaths(activeWorktree, liveWorktrees)
+  if (!activeWorktree) {
+    return paths
+  }
+  const setupsByRepoId = buildProjectSetupsByRepoId(options.projectHostSetupProjection)
+  for (const worktree of liveWorktrees) {
+    if (
+      worktree.repoId === activeWorktree.repoId ||
+      worktreeProjectKey(worktree) === options.activeProjectKey ||
+      (setupsByRepoId.get(worktree.repoId) ?? []).some(
+        (setup) => worktreeProjectKey(setup, setup) === options.activeProjectKey
+      )
+    ) {
+      addAiVaultWorkspaceScopePath(paths, worktree.path)
+    }
+  }
+  for (const setup of options.projectHostSetupProjection?.setups ?? []) {
+    if (worktreeProjectKey(setup, setup) === options.activeProjectKey) {
+      addAiVaultWorkspaceScopePath(paths, setup.path)
+    }
+  }
+  return paths
+}
+
+function buildProjectSetupsByRepoId(
+  projection?: ProjectHostSetupProjection
+): Map<string, ProjectHostSetupProjection['setups']> {
+  const setupsByRepoId = new Map<string, ProjectHostSetupProjection['setups']>()
+  for (const setup of projection?.setups ?? []) {
+    const setups = setupsByRepoId.get(setup.repoId) ?? []
+    setups.push(setup)
+    setupsByRepoId.set(setup.repoId, setups)
+  }
+  return setupsByRepoId
+}
+
+function worktreeProjectKey(
+  entry: Pick<Worktree, 'projectId' | 'repoId'> | { projectId?: string | null; repoId?: string },
+  setup?: { projectId?: string | null; repoId?: string }
+): string | null {
+  const projectId = entry.projectId ?? setup?.projectId ?? null
+  if (projectId) {
+    return projectId.startsWith('repo:') ? projectId : `project:${projectId}`
+  }
+  return entry.repoId ? `repo:${entry.repoId}` : null
+}
+
+function addAiVaultWorkspaceScopePath(paths: string[], pathValue: string): void {
+  const trimmedPath = pathValue.trim()
+  if (!trimmedPath || !isRuntimePathAbsolute(trimmedPath)) {
+    return
+  }
+  const comparisonPath = normalizeRuntimePathForComparison(trimmedPath)
+  if (
+    paths.some((existingPath) => normalizeRuntimePathForComparison(existingPath) === comparisonPath)
+  ) {
+    return
+  }
+  paths.push(trimmedPath)
+}
+
+function isAiVaultWorkspaceScopePathClaimed(
+  pathValue: string,
+  activeWorktree: Pick<Worktree, 'id'>,
+  liveWorktrees: readonly Pick<Worktree, 'id' | 'path'>[]
+): boolean {
+  const trimmedPath = pathValue.trim()
+  if (!trimmedPath || !isRuntimePathAbsolute(trimmedPath)) {
+    return false
+  }
+  const comparisonPath = normalizeRuntimePathForComparison(trimmedPath)
+  // AI Vault sessions are keyed by cwd only, so any live worktree now owning this path wins.
+  return liveWorktrees.some(
+    (worktree) =>
+      worktree.id !== activeWorktree.id &&
+      normalizeRuntimePathForComparison(worktree.path) === comparisonPath
+  )
+}

src/renderer/src/components/right-sidebar/ai-vault-session-filters.test.ts

@@ -2,13 +2,16 @@ import { describe, expect, it } from 'vitest'
 import type { AiVaultSession } from '../../../../shared/ai-vault-types'
 import {
   AI_VAULT_SESSION_FILTER_QUERY_MAX_BYTES,
-  deriveAiVaultWorkspaceScopePaths,
   filterAiVaultSessions,
   folderLabel,
   groupAiVaultSessions,
   isAiVaultSessionFilterQueryTooLarge,
   parseVaultQuery
 } from './ai-vault-session-filters'
+import {
+  deriveAiVaultScopeSessionPaths,
+  deriveAiVaultWorkspaceScopePaths
+} from './ai-vault-scope-paths'
 
 const baseSession: AiVaultSession = {
   id: 'claude:1',
@@ -415,6 +418,176 @@ describe('deriveAiVaultWorkspaceScopePaths', () => {
   })
 })
 
+describe('deriveAiVaultScopeSessionPaths', () => {
+  it('adds same-repo sibling worktrees on top of the workspace paths', () => {
+    expect(
+      deriveAiVaultScopeSessionPaths(
+        {
+          id: 'repo1::/Users/ada/workspaces/orca/fix-agent-history',
+          repoId: 'repo1',
+          path: '/Users/ada/workspaces/orca/fix-agent-history',
+          priorWorktreeIds: []
+        },
+        [
+          {
+            id: 'repo1::/Users/ada/workspaces/orca/fix-agent-history',
+            repoId: 'repo1',
+            path: '/Users/ada/workspaces/orca/fix-agent-history'
+          },
+          {
+            id: 'repo1::/Users/ada/workspaces/orca/sibling',
+            repoId: 'repo1',
+            path: '/Users/ada/workspaces/orca/sibling'
+          },
+          {
+            id: 'repo2::/Users/ada/workspaces/other/elsewhere',
+            repoId: 'repo2',
+            path: '/Users/ada/workspaces/other/elsewhere'
+          }
+        ]
+      )
+    ).toEqual([
+      '/Users/ada/workspaces/orca/fix-agent-history',
+      '/Users/ada/workspaces/orca/sibling'
+    ])
+  })
+
+  it('returns no paths without an active worktree', () => {
+    expect(deriveAiVaultScopeSessionPaths(null, [])).toEqual([])
+  })
+
+  it('adds active project setup paths across repos', () => {
+    expect(
+      deriveAiVaultScopeSessionPaths(
+        {
+          id: 'repo1::/Users/ada/workspaces/orca/app',
+          repoId: 'repo1',
+          path: '/Users/ada/workspaces/orca/app',
+          priorWorktreeIds: []
+        },
+        [
+          {
+            id: 'repo1::/Users/ada/workspaces/orca/app',
+            repoId: 'repo1',
+            path: '/Users/ada/workspaces/orca/app'
+          },
+          {
+            id: 'repo2::/Users/ada/workspaces/orca/docs',
+            repoId: 'repo2',
+            path: '/Users/ada/workspaces/orca/docs'
+          }
+        ],
+        {
+          activeProjectKey: 'project:orca',
+          projectHostSetupProjection: {
+            projects: [
+              {
+                id: 'orca',
+                displayName: 'Orca',
+                badgeColor: '#2563eb',
+                sourceRepoIds: ['repo1', 'repo2'],
+                createdAt: 1,
+                updatedAt: 1
+              }
+            ],
+            setups: [
+              {
+                id: 'setup-1',
+                projectId: 'orca',
+                hostId: 'local',
+                repoId: 'repo1',
+                displayName: 'App',
+                path: '/Users/ada/workspaces/orca/app',
+                setupState: 'ready',
+                setupMethod: 'imported-existing-folder',
+                createdAt: 1,
+                updatedAt: 1
+              },
+              {
+                id: 'setup-2',
+                projectId: 'orca',
+                hostId: 'local',
+                repoId: 'repo2',
+                displayName: 'Docs',
+                path: '/Users/ada/workspaces/orca/docs',
+                setupState: 'ready',
+                setupMethod: 'imported-existing-folder',
+                createdAt: 1,
+                updatedAt: 1
+              }
+            ]
+          }
+        }
+      )
+    ).toEqual(['/Users/ada/workspaces/orca/app', '/Users/ada/workspaces/orca/docs'])
+  })
+
+  it('keeps live worktree paths when another setup shares the repo id', () => {
+    expect(
+      deriveAiVaultScopeSessionPaths(
+        {
+          id: 'repo1::/Users/ada/workspaces/orca/app',
+          repoId: 'repo1',
+          path: '/Users/ada/workspaces/orca/app',
+          priorWorktreeIds: []
+        },
+        [
+          {
+            id: 'repo2::/Users/ada/workspaces/orca/docs-worktree',
+            repoId: 'repo2',
+            path: '/Users/ada/workspaces/orca/docs-worktree'
+          }
+        ],
+        {
+          activeProjectKey: 'project:orca',
+          projectHostSetupProjection: {
+            projects: [
+              {
+                id: 'orca',
+                displayName: 'Orca',
+                badgeColor: '#2563eb',
+                sourceRepoIds: ['repo1', 'repo2'],
+                createdAt: 1,
+                updatedAt: 1
+              }
+            ],
+            setups: [
+              {
+                id: 'setup-1',
+                projectId: 'orca',
+                hostId: 'local',
+                repoId: 'repo2',
+                displayName: 'Docs',
+                path: '/Users/ada/workspaces/orca/docs',
+                setupState: 'ready',
+                setupMethod: 'imported-existing-folder',
+                createdAt: 1,
+                updatedAt: 1
+              },
+              {
+                id: 'setup-2',
+                projectId: 'other',
+                hostId: 'local',
+                repoId: 'repo2',
+                displayName: 'Other',
+                path: '/Users/ada/workspaces/other',
+                setupState: 'ready',
+                setupMethod: 'imported-existing-folder',
+                createdAt: 1,
+                updatedAt: 1
+              }
+            ]
+          }
+        }
+      )
+    ).toEqual([
+      '/Users/ada/workspaces/orca/app',
+      '/Users/ada/workspaces/orca/docs-worktree',
+      '/Users/ada/workspaces/orca/docs'
+    ])
+  })
+})
+
 describe('isAiVaultSessionFilterQueryTooLarge', () => {
   it('counts UTF-8 bytes rather than UTF-16 code units', () => {
     expect(

src/renderer/src/components/right-sidebar/ai-vault-session-filters.ts

@@ -1,7 +1,5 @@
 import {
   isPathInsideOrEqual,
-  isRuntimePathAbsolute,
-  normalizeRuntimePathForComparison,
   normalizeRuntimePathSeparators
 } from '../../../../shared/cross-platform-path'
 import { isClipboardTextByteLengthOverLimit } from '../../../../shared/clipboard-text'
@@ -14,8 +12,6 @@ import type {
   AiVaultSort
 } from '../../../../shared/ai-vault-types'
 import { aiVaultAgentLabel } from '../../../../shared/ai-vault-types'
-import type { Worktree } from '../../../../shared/types'
-import { splitWorktreeIdForFilesystem } from '../../../../shared/worktree-id'
 import { sessionPreviewSearchText } from './ai-vault-session-display'
 import type { AiVaultSessionProject } from './ai-vault-session-projects'
 
@@ -95,31 +91,6 @@ export function filterAiVaultSessions(
     .sort((left, right) => compareSessions(left, right, filters.sort))
 }
 
-export function deriveAiVaultWorkspaceScopePaths(
-  activeWorktree: Pick<Worktree, 'id' | 'path' | 'priorWorktreeIds' | 'repoId'> | null,
-  liveWorktrees: readonly Pick<Worktree, 'id' | 'path' | 'repoId'>[] = []
-): string[] {
-  if (!activeWorktree) {
-    return []
-  }
-
-  const paths: string[] = []
-  addAiVaultWorkspaceScopePath(paths, activeWorktree.path)
-
-  for (const priorWorktreeId of activeWorktree.priorWorktreeIds ?? []) {
-    const parsed = splitWorktreeIdForFilesystem(priorWorktreeId)
-    if (!parsed || parsed.repoId !== activeWorktree.repoId) {
-      continue
-    }
-    if (isAiVaultWorkspaceScopePathClaimed(parsed.worktreePath, activeWorktree, liveWorktrees)) {
-      continue
-    }
-    addAiVaultWorkspaceScopePath(paths, parsed.worktreePath)
-  }
-
-  return paths
-}
-
 export function groupAiVaultSessions(
   sessions: readonly AiVaultSession[],
   group: AiVaultGroup,
@@ -264,20 +235,6 @@ function getFolderGroupKey(pathValue: string | null): string {
   return pathValue ? normalizeRuntimePathSeparators(pathValue).toLowerCase() : 'unknown'
 }
 
-function addAiVaultWorkspaceScopePath(paths: string[], pathValue: string): void {
-  const trimmedPath = pathValue.trim()
-  if (!trimmedPath || !isRuntimePathAbsolute(trimmedPath)) {
-    return
-  }
-  const comparisonPath = normalizeRuntimePathForComparison(trimmedPath)
-  if (
-    paths.some((existingPath) => normalizeRuntimePathForComparison(existingPath) === comparisonPath)
-  ) {
-    return
-  }
-  paths.push(trimmedPath)
-}
-
 function isAiVaultSessionInWorkspacePath(workspacePath: string, sessionCwd: string): boolean {
   if (isPathInsideOrEqual(workspacePath, sessionCwd)) {
     return true
@@ -293,24 +250,6 @@ function isAiVaultSessionInWorkspacePath(workspacePath: string, sessionCwd: stri
   return isPathInsideOrEqual(workspaceWslPath.linuxPath, sessionCwd)
 }
 
-function isAiVaultWorkspaceScopePathClaimed(
-  pathValue: string,
-  activeWorktree: Pick<Worktree, 'id'>,
-  liveWorktrees: readonly Pick<Worktree, 'id' | 'path'>[]
-): boolean {
-  const trimmedPath = pathValue.trim()
-  if (!trimmedPath || !isRuntimePathAbsolute(trimmedPath)) {
-    return false
-  }
-  const comparisonPath = normalizeRuntimePathForComparison(trimmedPath)
-  // AI Vault sessions are keyed by cwd only, so any live worktree now owning this path wins.
-  return liveWorktrees.some(
-    (worktree) =>
-      worktree.id !== activeWorktree.id &&
-      normalizeRuntimePathForComparison(worktree.path) === comparisonPath
-  )
-}
-
 function tokenizeQuery(query: string): string[] {
   const tokens: string[] = []
   const pattern = /"([^"]+)"|'([^']+)'|(\S+)/g

src/renderer/src/components/right-sidebar/ai-vault-session-refresh.ts

@@ -0,0 +1,87 @@
+import { useCallback, useEffect, useMemo, useRef, useState } from 'react'
+import type { AiVaultListResult, AiVaultSession } from '../../../../shared/ai-vault-types'
+
+const SESSION_LIMIT = 500
+
+export function useAiVaultSessionRefresh(scopePaths: readonly string[]): {
+  error: string | null
+  loading: boolean
+  refresh: (args?: { force?: boolean }) => Promise<void>
+  scanResult: AiVaultListResult | null
+  sessions: AiVaultSession[]
+} {
+  const [sessions, setSessions] = useState<AiVaultSession[]>([])
+  const [scanResult, setScanResult] = useState<AiVaultListResult | null>(null)
+  const [loading, setLoading] = useState(false)
+  const [error, setError] = useState<string | null>(null)
+  const refreshIdRef = useRef(0)
+  const refreshInFlightRef = useRef(false)
+  const pendingRefreshRef = useRef(false)
+  const pendingForceRef = useRef(false)
+  const mountedRef = useRef(true)
+  const scopePathsKey = useMemo(() => scopePaths.join('\n'), [scopePaths])
+  const scopePathsRef = useRef<readonly string[]>(scopePaths)
+  scopePathsRef.current = scopePaths
+
+  const refresh = useCallback(async (args: { force?: boolean } = {}): Promise<void> => {
+    // A scope change during an in-flight scan must not be dropped; queue one more
+    // scan so the current scoped view is refreshed after the older scan settles.
+    if (refreshInFlightRef.current) {
+      pendingRefreshRef.current = true
+      pendingForceRef.current ||= args.force === true
+      return
+    }
+
+    refreshInFlightRef.current = true
+    const refreshId = refreshIdRef.current + 1
+    refreshIdRef.current = refreshId
+    setLoading(true)
+    setError(null)
+    try {
+      const result = await window.api.aiVault.listSessions({
+        limit: SESSION_LIMIT,
+        scopePaths: scopePathsRef.current,
+        force: args.force
+      })
+      if (!mountedRef.current || refreshIdRef.current !== refreshId) {
+        return
+      }
+      setScanResult(result)
+      setSessions(result.sessions)
+    } catch (err) {
+      if (mountedRef.current && refreshIdRef.current === refreshId) {
+        setError(err instanceof Error ? err.message : String(err))
+      }
+    } finally {
+      refreshInFlightRef.current = false
+      if (mountedRef.current && refreshIdRef.current === refreshId) {
+        setLoading(false)
+      }
+      if (pendingRefreshRef.current && mountedRef.current) {
+        pendingRefreshRef.current = false
+        const force = pendingForceRef.current
+        pendingForceRef.current = false
+        void refresh({ force })
+      }
+    }
+    // Deps are intentionally empty: refresh reads changing values through refs
+    // and recurses on itself, so its identity must stay stable.
+  }, [])
+
+  useEffect(() => {
+    mountedRef.current = true
+    return () => {
+      mountedRef.current = false
+      refreshIdRef.current += 1
+      refreshInFlightRef.current = false
+    }
+  }, [])
+
+  // Re-scan on mount and whenever the active scope changes, since the scanner
+  // tailors its in-scope results to scopePaths.
+  useEffect(() => {
+    void refresh()
+  }, [refresh, scopePathsKey])
+
+  return { error, loading, refresh, scanResult, sessions }
+}

src/shared/ai-vault-types.ts

@@ -74,6 +74,9 @@ export type AiVaultScanIssue = {
 export type AiVaultListArgs = {
   limit?: number
   force?: boolean
+  // Active workspace/project paths. The global result is recency-capped, so these
+  // guarantee a scoped view still surfaces its own (possibly older) sessions.
+  scopePaths?: readonly string[]
 }
 
 export type AiVaultListResult = {