typeclaw 0.45.1 → 0.46.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (142) hide show
  1. package/README.md +1 -1
  2. package/package.json +2 -1
  3. package/src/agent/git-nudge.ts +2 -1
  4. package/src/agent/index.ts +44 -30
  5. package/src/agent/live-subagents.ts +18 -0
  6. package/src/agent/multimodal/index.ts +1 -1
  7. package/src/agent/multimodal/look-at.ts +78 -44
  8. package/src/agent/multimodal/looker.ts +151 -39
  9. package/src/agent/network/response-body.ts +26 -0
  10. package/src/agent/network/safe-http.ts +198 -0
  11. package/src/agent/plugin-tools.ts +379 -157
  12. package/src/agent/provider-error.ts +52 -1
  13. package/src/agent/restart/index.ts +61 -40
  14. package/src/agent/restart-handoff/index.ts +73 -4
  15. package/src/agent/reviewer-bash-policy.ts +42 -1
  16. package/src/agent/session-origin.ts +2 -0
  17. package/src/agent/subagent-drain.ts +117 -7
  18. package/src/agent/subagents.ts +49 -16
  19. package/src/agent/throttle-circuit.ts +25 -5
  20. package/src/agent/tool-error-remediation.ts +72 -0
  21. package/src/agent/tool-file-safety.ts +1050 -0
  22. package/src/agent/tools/channel-fetch-attachment.ts +10 -3
  23. package/src/agent/tools/ddg.ts +16 -2
  24. package/src/agent/tools/post-github-review.ts +199 -0
  25. package/src/agent/tools/spawn-subagent.ts +21 -18
  26. package/src/agent/tools/webfetch/fetch.ts +143 -183
  27. package/src/agent/tools/webfetch/tool.ts +2 -3
  28. package/src/agent/tools/websearch.ts +23 -2
  29. package/src/agent/tools-without-local-file-operands.ts +27 -0
  30. package/src/agent/turn-runner.ts +86 -6
  31. package/src/bundled-plugins/backup/index.ts +2 -1
  32. package/src/bundled-plugins/backup/runner.ts +2 -1
  33. package/src/bundled-plugins/doc-render/render.ts +93 -12
  34. package/src/bundled-plugins/doc-render/skills/typeclaw-render-pdf/SKILL.md +13 -13
  35. package/src/bundled-plugins/github-cli-auth/effective-approval.ts +16 -6
  36. package/src/bundled-plugins/github-cli-auth/gh-command.ts +555 -15
  37. package/src/bundled-plugins/github-cli-auth/index.ts +131 -135
  38. package/src/bundled-plugins/github-cli-auth/review-checkout.ts +138 -0
  39. package/src/bundled-plugins/guard/policies/uncommitted-changes.ts +2 -1
  40. package/src/bundled-plugins/memory/README.md +23 -11
  41. package/src/bundled-plugins/memory/append-tool.ts +15 -8
  42. package/src/bundled-plugins/memory/citations.ts +22 -9
  43. package/src/bundled-plugins/memory/dreaming.ts +13 -8
  44. package/src/bundled-plugins/memory/find-entry-tool.ts +4 -0
  45. package/src/bundled-plugins/memory/index.ts +0 -4
  46. package/src/bundled-plugins/memory/load-memory.ts +8 -3
  47. package/src/bundled-plugins/memory/provenance-index.ts +717 -0
  48. package/src/bundled-plugins/memory/provenance-sanitize.ts +40 -0
  49. package/src/bundled-plugins/memory/references/store-reference-tool.ts +4 -0
  50. package/src/bundled-plugins/memory/search-tool.ts +163 -60
  51. package/src/bundled-plugins/memory/stream-events.ts +2 -0
  52. package/src/bundled-plugins/memory/vector/hybrid.ts +126 -13
  53. package/src/bundled-plugins/researcher/write-report.ts +6 -3
  54. package/src/bundled-plugins/reviewer/index.ts +2 -2
  55. package/src/bundled-plugins/reviewer/reviewer.ts +10 -3
  56. package/src/bundled-plugins/reviewer/skills/code-review.ts +19 -9
  57. package/src/bundled-plugins/scout/scout.ts +8 -0
  58. package/src/bundled-plugins/security/index.ts +4 -4
  59. package/src/bundled-plugins/security/policies/outbound-secret-scan.ts +1 -0
  60. package/src/bundled-plugins/security/policies/private-surface-read.ts +220 -29
  61. package/src/bundled-plugins/security/policies/secret-exfil-read.ts +1 -0
  62. package/src/bundled-plugins/security/policies/ssrf.ts +29 -9
  63. package/src/channels/adapters/discord-bot-channel-resolver.ts +26 -14
  64. package/src/channels/adapters/discord-bot-thread-room.ts +44 -12
  65. package/src/channels/adapters/discord-bot.ts +66 -6
  66. package/src/channels/adapters/discord-channel-resolver.ts +91 -23
  67. package/src/channels/adapters/discord-id.ts +4 -0
  68. package/src/channels/adapters/discord-resolver-cache.ts +41 -0
  69. package/src/channels/adapters/discord.ts +57 -7
  70. package/src/channels/adapters/github/auth-app.ts +41 -12
  71. package/src/channels/adapters/github/index.ts +23 -11
  72. package/src/channels/adapters/github/review-submitter.ts +302 -0
  73. package/src/channels/adapters/kakaotalk-channel-resolver.ts +5 -3
  74. package/src/channels/adapters/kakaotalk-fetch-attachment.ts +8 -4
  75. package/src/channels/adapters/kakaotalk-typing.ts +17 -15
  76. package/src/channels/adapters/slack-attachment-download.ts +71 -0
  77. package/src/channels/adapters/slack-bot.ts +19 -8
  78. package/src/channels/adapters/slack-classify.ts +21 -2
  79. package/src/channels/adapters/slack.ts +59 -7
  80. package/src/channels/adapters/telegram-bot.ts +18 -5
  81. package/src/channels/adapters/webex-bot.ts +8 -3
  82. package/src/channels/adapters/webex.ts +8 -3
  83. package/src/channels/fetch-attachment.ts +43 -0
  84. package/src/{bundled-plugins/github-cli-auth/approve-idempotency.ts → channels/github-review-verdict-coordinator.ts} +25 -1
  85. package/src/channels/github-token-bridge.ts +10 -3
  86. package/src/channels/manager.ts +9 -0
  87. package/src/channels/router.ts +248 -6
  88. package/src/channels/types.ts +28 -1
  89. package/src/cli/channel.ts +1 -1
  90. package/src/cli/init.ts +1 -1
  91. package/src/cli/model.ts +57 -7
  92. package/src/compose/stop.ts +17 -5
  93. package/src/config/config.ts +6 -6
  94. package/src/config/models-mutation.ts +13 -1
  95. package/src/config/providers.ts +176 -0
  96. package/src/container/start.ts +49 -10
  97. package/src/doctor/checks.ts +3 -1
  98. package/src/doctor/commit.ts +2 -1
  99. package/src/doctor/file-ownership.ts +52 -0
  100. package/src/dreams/git.ts +2 -1
  101. package/src/git/hookless.ts +5 -0
  102. package/src/git/reconcile-ignored.ts +18 -5
  103. package/src/git/secret-history.ts +371 -0
  104. package/src/git/system-commit.ts +12 -6
  105. package/src/init/dockerfile.ts +86 -5
  106. package/src/init/index.ts +17 -5
  107. package/src/init/models-dev.ts +4 -0
  108. package/src/init/validate-api-key.ts +2 -0
  109. package/src/mcp/tools.ts +16 -3
  110. package/src/permissions/builtins.ts +3 -1
  111. package/src/plugin/context.ts +2 -0
  112. package/src/plugin/index.ts +1 -0
  113. package/src/plugin/manager.ts +4 -0
  114. package/src/plugin/types.ts +21 -0
  115. package/src/run/index.ts +40 -0
  116. package/src/run/llm-fetch-observer.ts +41 -4
  117. package/src/sandbox/build.ts +37 -15
  118. package/src/sandbox/canonical-secrets.ts +18 -0
  119. package/src/sandbox/env-exposure.ts +80 -0
  120. package/src/sandbox/errors.ts +13 -6
  121. package/src/sandbox/hidden-paths.ts +102 -44
  122. package/src/sandbox/index.ts +22 -4
  123. package/src/sandbox/package-install.ts +3 -27
  124. package/src/sandbox/policy.ts +11 -25
  125. package/src/sandbox/privileged-runtime.ts +255 -0
  126. package/src/sandbox/session-tmp.ts +8 -2
  127. package/src/sandbox/writable-zones.ts +204 -140
  128. package/src/secrets/index.ts +8 -0
  129. package/src/secrets/refresh-provider-oauth.ts +128 -0
  130. package/src/server/command-runner.ts +48 -4
  131. package/src/server/index.ts +4 -0
  132. package/src/skills/typeclaw-channel-github/SKILL.md +49 -56
  133. package/src/skills/typeclaw-claude-code/SKILL.md +24 -60
  134. package/src/skills/typeclaw-claude-code/references/auth-flow.md +17 -158
  135. package/src/skills/typeclaw-codex-cli/SKILL.md +19 -49
  136. package/src/skills/typeclaw-codex-cli/references/auth-flow.md +17 -154
  137. package/src/skills/typeclaw-config/SKILL.md +5 -5
  138. package/src/skills/typeclaw-config/references/recommended-mounts.md +24 -24
  139. package/src/skills/typeclaw-github-contributing/SKILL.md +23 -26
  140. package/src/skills/typeclaw-memory/SKILL.md +9 -3
  141. package/src/skills/typeclaw-permissions/SKILL.md +19 -15
  142. package/typeclaw.schema.json +20 -0
@@ -0,0 +1,40 @@
1
+ import { detectSecrets } from './secret-detector'
2
+
3
+ export const MAX_PROVENANCE_NAME_LENGTH = 256
4
+
5
+ export function sanitizeProvenanceName(value: string | undefined): string | undefined {
6
+ if (value === undefined || value.length === 0 || value.length > MAX_PROVENANCE_NAME_LENGTH) return undefined
7
+ if (hasUnsafeUnicode(value) || hasMarkdownPromptShaping(value) || detectSecrets(value).length > 0) return undefined
8
+ return value
9
+ }
10
+
11
+ export function isSafeProvenanceCoordinate(value: string): boolean {
12
+ return value.length > 0 && value.length <= MAX_PROVENANCE_NAME_LENGTH && !hasUnsafeUnicode(value)
13
+ }
14
+
15
+ function hasUnsafeUnicode(value: string): boolean {
16
+ for (const character of value) {
17
+ const codePoint = character.codePointAt(0)
18
+ if (codePoint === undefined) continue
19
+ if (codePoint <= 0x1f || (codePoint >= 0x7f && codePoint <= 0x9f)) return true
20
+ if (codePoint === 0x00ad || codePoint === 0x034f || codePoint === 0x180e) return true
21
+ if (codePoint === 0x061c || codePoint === 0x200e || codePoint === 0x200f) return true
22
+ if (codePoint >= 0x2028 && codePoint <= 0x202e) return true
23
+ if (codePoint >= 0x2061 && codePoint <= 0x2064) return true
24
+ if (codePoint >= 0x2066 && codePoint <= 0x2069) return true
25
+ if (codePoint === 0x200b || codePoint === 0x200c || codePoint === 0x200d) return true
26
+ if (codePoint === 0x2060 || codePoint === 0xfeff) return true
27
+ if (codePoint >= 0xe0000 && codePoint <= 0xe007f) return true
28
+ }
29
+ return false
30
+ }
31
+
32
+ function hasMarkdownPromptShaping(value: string): boolean {
33
+ if (['`', '*', '_', '[', ']', '<', '>', '|', '\\'].some((character) => value.includes(character))) return true
34
+ if (value.includes('__') || value.includes('![')) return true
35
+ const trimmed = value.trimStart()
36
+ if (trimmed.startsWith('#') || trimmed.startsWith('>') || trimmed.startsWith('---')) return true
37
+ if (trimmed.startsWith('- ') || trimmed.startsWith('+ ')) return true
38
+ if (/^\d{1,9}[.)]\s/u.test(trimmed)) return true
39
+ return false
40
+ }
@@ -23,6 +23,10 @@ export function createStoreReferenceTool(onReferenceStored?: ReferenceStoredHook
23
23
  origin: z.enum(['episode', 'curated', 'external']),
24
24
  tags: z.array(z.string()).optional(),
25
25
  }),
26
+ // title/body/tags are prose the tool writes under a runtime-computed slug
27
+ // path; none is a caller-supplied file path. Declared so a title like
28
+ // "README.md" or a body carrying "/" is not misread as a file operand.
29
+ fileOperands: { nonFile: ['title', 'body', 'tags'] },
26
30
  async execute({ title, body, origin, tags }, ctx) {
27
31
  const existingSlugs = new Set(await listReferenceSlugs(ctx.agentDir))
28
32
  const slug = headingToSlug(title, existingSlugs)
@@ -5,10 +5,17 @@ import { z } from 'zod'
5
5
  import { defineTool } from '@/plugin'
6
6
 
7
7
  import { loadAllShards, loadShard, type TopicShard } from './load-shards'
8
+ import {
9
+ buildProvenanceIndex,
10
+ buildProvenanceIndexFrom,
11
+ type MemoryScope,
12
+ type ProvenanceChild,
13
+ } from './provenance-index'
14
+ import { sanitizeProvenanceName } from './provenance-sanitize'
8
15
  import { renderReference } from './references/frontmatter'
9
16
  import { loadAllReferences, loadReference, type Reference } from './references/load-references'
10
17
  import type { FragmentEvent, FragmentProvenance, LegacyProseEvent, StreamEvent } from './stream-events'
11
- import { readAllUndreamedStreamDays, type UndreamedStreamDay } from './stream-io'
18
+ import { filterUndreamedEvents, readAllStreamDays, type UndreamedStreamDay } from './stream-io'
12
19
 
13
20
  const DEFAULT_MAX_RESULTS = 10
14
21
  const EXCERPT_CONTEXT_LINES = 3
@@ -21,6 +28,7 @@ export type TopicMatch = {
21
28
  heading: string
22
29
  excerpt: string
23
30
  fullBody?: string
31
+ provenance?: ProvenanceChild[]
24
32
  }
25
33
 
26
34
  export type StreamMatch = {
@@ -54,7 +62,7 @@ export type Matcher = (haystack: string) => boolean
54
62
  export function createMemorySearchTool() {
55
63
  return defineTool({
56
64
  description:
57
- 'Search the agent\'s long-term memory, or look up one topic shard by exact slug. Covers topic shards under memory/topics/ (consolidated facts), references under memory/references/ (verbatim artifacts), and undreamed daily-stream events under memory/streams/ (recent fragments not yet folded into shards). Pass `query` for search OR `topic` for an exact slug lookup, not both. Search is case-insensitive substring by default: tries the whole query as one phrase first, and if that finds nothing, falls back to OR-matching the individual words (ranked by how many words each hit contains) — so a multi-word query still returns results even when no entry contains the exact phrase. asRegex=true treats query as a JavaScript regex (no word fallback). `topic` skips search entirely and returns that one shard (or reference) with its full body — use it to read a topic OR reference whose slug you already have (e.g. a heading shown in injected memory); it resolves the topic shard first and falls back to a reference of the same slug. Returns matches discriminated by `source: "topic" | "reference" | "stream"`, each with line-context excerpts; full=true includes complete bodies (topic lookups always include the full body). Ordering depends on mode: exact-phrase (and regex) results list all topic matches first (alphabetical by slug), then reference matches, then stream matches (newest day first); word-fallback results are ranked by matched-word count, with that same topic-then-reference-then-stream-newest order as the tiebreak within each score band, so a higher-scoring stream match can precede a lower-scoring topic match. Pass `where` (a channel/room name like "#incidents" or its raw chat id) to scope the search to one room — use it for questions like "what did we discuss in #incidents"; it matches a stream fragment by its captured room id or human-readable name (case-insensitive, leading "#" optional) and, because room is fragment-only provenance, returns ONLY stream matches (topic shards and references carry no room).',
65
+ 'Search long-term memory and recent undreamed fragments, or look up one exact topic/reference slug. Plain search is case-insensitive phrase-first with token-OR fallback; regex mode has no token fallback. Runtime-owned provenance from active topic citations is searchable but never embedded. `workspace`, `chat`, and `thread` scope both topics and recent fragments to matching captured origin IDs or names; legacy `where` remains an alias for `chat`. Scoped topic eligibility, lexical matching, and returned provenance use the same eligible active cited children. Unresolved citations remain explicit in unscoped results and never gain invented metadata.',
58
66
  parameters: z.object({
59
67
  query: z.string().optional(),
60
68
  topic: z.string().optional(),
@@ -64,47 +72,95 @@ export function createMemorySearchTool() {
64
72
  since: z.string().optional(),
65
73
  before: z.string().optional(),
66
74
  where: z.string().optional(),
75
+ workspace: z.string().optional(),
76
+ chat: z.string().optional(),
77
+ thread: z.string().optional(),
67
78
  }),
68
- async execute({ query, topic, asRegex, full, maxResults, since, before, where }, ctx) {
79
+ // Every string field is a search term, an exact topic/reference slug, an ISO
80
+ // date, or an origin id/name — none is a local path (the tool reads memory/
81
+ // internally, never a caller path). Declared so a `topic` slug that collides
82
+ // with an agent-root dir, or a `since` date like "2026-01-01", is not
83
+ // misread as a file operand.
84
+ fileOperands: {
85
+ nonFile: ['query', 'topic', 'since', 'before', 'where', 'workspace', 'chat', 'thread'],
86
+ },
87
+ async execute({ query, topic, asRegex, full, maxResults, since, before, where, workspace, chat, thread }, ctx) {
69
88
  if ((query === undefined) === (topic === undefined)) {
70
89
  return resultToToolResult({ error: 'provide exactly one of `query` or `topic`' })
71
90
  }
72
91
 
92
+ const chatScope = chat ?? where
93
+ const requestedScope: MemoryScope = {
94
+ ...(workspace === undefined ? {} : { workspace }),
95
+ ...(chatScope === undefined ? {} : { chat: chatScope }),
96
+ ...(thread === undefined ? {} : { thread }),
97
+ }
98
+ const scope = requestedScope
99
+
73
100
  if (topic !== undefined) {
74
- return resultToToolResult(await lookupTopic(ctx.agentDir, topic, ctx.logger))
101
+ return resultToToolResult(await lookupTopic(ctx.agentDir, topic, scope, ctx.logger))
75
102
  }
76
103
 
77
104
  const matcherOrError = buildMatcher(query!, asRegex)
78
105
  if (typeof matcherOrError === 'string') {
79
106
  return resultToToolResult({ error: matcherOrError })
80
107
  }
81
-
82
- const [shards, streamDays, allReferences] = await Promise.all([
108
+ const [shards, allStreamDays, allReferences] = await Promise.all([
83
109
  loadAllShards(ctx.agentDir, { logger: ctx.logger }),
84
- readAllUndreamedStreamDays(ctx.agentDir),
110
+ readAllStreamDays(ctx.agentDir),
85
111
  loadAllReferences(ctx.agentDir, { logger: ctx.logger }),
86
112
  ])
113
+ const provenanceIndex = await buildProvenanceIndexFrom(ctx.agentDir, shards, allStreamDays)
114
+ const scoped = hasMemoryScope(scope)
115
+ const streamDays: UndreamedStreamDay[] = allStreamDays.flatMap((day) => {
116
+ const events: StreamEvent[] = []
117
+ for (const event of filterUndreamedEvents(day.events, day.dreamedIds)) {
118
+ if (event.type !== 'fragment') {
119
+ if (!scoped) events.push(event)
120
+ continue
121
+ }
122
+ const citation = `streams/${day.date}#${event.id}`
123
+ if (provenanceIndex.isActivelyCited(citation) || provenanceIndex.isSuperseded(citation)) continue
124
+ const child = provenanceIndex.undreamedChild(citation, scope)
125
+ if (child !== undefined) events.push(safeStreamEvent(event, child))
126
+ }
127
+ return events.length === 0 ? [] : [{ date: day.date, path: day.path, name: day.name, events }]
128
+ })
87
129
  const dateFilter = parseReferenceDateFilter(since, before)
88
130
  if ('error' in dateFilter) return resultToToolResult(dateFilter)
89
131
 
90
- // A `where` filter scopes the search to ONE room. `where` is fragment-only
91
- // provenance, so a room query excludes topic shards and references (they
92
- // carry no room) and keeps only stream fragments whose `where` matches.
93
- const roomFilteredDays = where === undefined ? streamDays : filterStreamDaysByRoom(streamDays, where)
94
- const scopedShards = where === undefined ? shards : []
95
- const references =
96
- where === undefined ? allReferences.filter((r) => referenceCandidateAllowed(r, dateFilter)) : []
132
+ const roomFilteredDays = streamDays
133
+ const scopedShards = shards.filter((shard) => provenanceIndex.topicEligible(shard.slug, scope))
134
+ const references = !scoped ? allReferences.filter((r) => referenceCandidateAllowed(r, dateFilter)) : []
135
+ const topicDocuments = new Map(
136
+ scopedShards.map((shard) => [
137
+ shard.slug,
138
+ `${shardSearchText(shard)}\n${provenanceIndex.lexicalTextForTopic(shard.slug, scope)}`,
139
+ ]),
140
+ )
141
+ const topicProvenance = new Map(
142
+ scopedShards.map((shard) => [shard.slug, provenanceIndex.childrenForTopic(shard.slug, scope)]),
143
+ )
144
+ const streamDocuments = new Map<string, string>()
145
+ for (const day of roomFilteredDays) {
146
+ for (const event of day.events) {
147
+ if (event.type !== 'fragment') continue
148
+ const citation = `streams/${day.date}#${event.id}`
149
+ streamDocuments.set(
150
+ citation,
151
+ `${eventSearchText(event)}\n${provenanceIndex.lexicalTextForUndreamed(citation, scope)}`,
152
+ )
153
+ }
154
+ }
97
155
 
98
156
  if (scopedShards.length === 0 && roomFilteredDays.length === 0 && references.length === 0) {
99
157
  return resultToToolResult({ matches: [], truncatedAt: 0 })
100
158
  }
101
159
 
102
- let result = searchAll(scopedShards, roomFilteredDays, matcherOrError, { full, maxResults, references })
160
+ const searchOptions = { full, maxResults, references, topicDocuments, topicProvenance, streamDocuments }
161
+ let result = searchAll(scopedShards, roomFilteredDays, matcherOrError, searchOptions)
103
162
  if ('matches' in result && result.matches.length === 0) {
104
- const fallback = tokenFallback(query!, asRegex, scopedShards, roomFilteredDays, references, {
105
- full,
106
- maxResults,
107
- })
163
+ const fallback = tokenFallback(query!, asRegex, scopedShards, roomFilteredDays, references, searchOptions)
108
164
  if (fallback !== null) result = fallback
109
165
  }
110
166
  if ('matches' in result) await bumpReturnedReferences(allReferences, result.matches)
@@ -113,6 +169,15 @@ export function createMemorySearchTool() {
113
169
  })
114
170
  }
115
171
 
172
+ function safeStreamEvent(event: FragmentEvent, child: ProvenanceChild): FragmentEvent {
173
+ const safe = { ...event }
174
+ delete safe.who
175
+ delete safe.where
176
+ if (child.who !== undefined) safe.who = child.who
177
+ if (child.where !== undefined) safe.where = child.where
178
+ return safe
179
+ }
180
+
116
181
  export const memorySearchTool = createMemorySearchTool()
117
182
 
118
183
  // Exact slug lookup, so the agent can read a topic OR reference whose slug the
@@ -127,6 +192,7 @@ export const memorySearchTool = createMemorySearchTool()
127
192
  async function lookupTopic(
128
193
  agentDir: string,
129
194
  slug: string,
195
+ scope: MemoryScope,
130
196
  logger?: { warn(message: string): void },
131
197
  ): Promise<MemorySearchResult> {
132
198
  const loaderOptions = logger === undefined ? {} : { logger }
@@ -136,8 +202,14 @@ async function lookupTopic(
136
202
  } catch (err) {
137
203
  return { error: `invalid topic slug: ${err instanceof Error ? err.message : String(err)}` }
138
204
  }
139
- if (shard !== null) return { matches: [topicMatchWithFullBody(shard)] }
205
+ if (shard !== null) {
206
+ const index = await buildProvenanceIndex(agentDir)
207
+ if (!index.topicEligible(shard.slug, scope)) return { matches: [] }
208
+ const provenance = index.childrenForTopic(shard.slug, scope)
209
+ return { matches: [topicMatchWithFullBody(shard, provenance)] }
210
+ }
140
211
 
212
+ if (hasMemoryScope(scope)) return { matches: [] }
141
213
  const reference = await loadReference(agentDir, slug, loaderOptions)
142
214
  if (reference !== null) {
143
215
  const match = referenceMatchWithFullBody(reference)
@@ -151,7 +223,11 @@ async function lookupTopic(
151
223
  return { matches: [] }
152
224
  }
153
225
 
154
- function topicMatchWithFullBody(shard: TopicShard): TopicMatch {
226
+ function hasMemoryScope(scope: MemoryScope): boolean {
227
+ return scope.workspace !== undefined || scope.chat !== undefined || scope.thread !== undefined
228
+ }
229
+
230
+ function topicMatchWithFullBody(shard: TopicShard, provenance: ProvenanceChild[] = []): TopicMatch {
155
231
  return {
156
232
  source: 'topic',
157
233
  shardPath: shard.path,
@@ -159,6 +235,7 @@ function topicMatchWithFullBody(shard: TopicShard): TopicMatch {
159
235
  heading: shard.frontmatter.heading,
160
236
  excerpt: excerpt(shard.body),
161
237
  fullBody: shard.body,
238
+ ...(provenance.length > 0 ? { provenance } : {}),
162
239
  }
163
240
  }
164
241
 
@@ -193,7 +270,13 @@ function tokenFallback(
193
270
  shards: TopicShard[],
194
271
  streamDays: UndreamedStreamDay[],
195
272
  references: Reference[],
196
- options: { full: boolean; maxResults: number },
273
+ options: {
274
+ full: boolean
275
+ maxResults: number
276
+ topicDocuments?: Map<string, string>
277
+ topicProvenance?: Map<string, ProvenanceChild[]>
278
+ streamDocuments?: Map<string, string>
279
+ },
197
280
  ): MemorySearchResult | null {
198
281
  if (asRegex) return null
199
282
  const tokens = distinctTokens(query)
@@ -238,7 +321,14 @@ export function searchAll(
238
321
  shards: TopicShard[],
239
322
  streamDays: UndreamedStreamDay[],
240
323
  matcher: Matcher,
241
- options: { full: boolean; maxResults: number; references?: Reference[] },
324
+ options: {
325
+ full: boolean
326
+ maxResults: number
327
+ references?: Reference[]
328
+ topicDocuments?: Map<string, string>
329
+ topicProvenance?: Map<string, ProvenanceChild[]>
330
+ streamDocuments?: Map<string, string>
331
+ },
242
332
  ): MemorySearchResult {
243
333
  const matches: MemorySearchMatch[] = []
244
334
  let truncatedAt: number | undefined
@@ -253,7 +343,13 @@ export function searchAll(
253
343
  }
254
344
 
255
345
  for (const shard of shards) {
256
- const match = matchShard(shard, matcher, options.full)
346
+ const match = matchShard(
347
+ shard,
348
+ matcher,
349
+ options.full,
350
+ options.topicDocuments?.get(shard.slug),
351
+ options.topicProvenance?.get(shard.slug),
352
+ )
257
353
  if (match === null) continue
258
354
  if (!push(match)) return { matches, truncatedAt: truncatedAt! }
259
355
  }
@@ -267,7 +363,7 @@ export function searchAll(
267
363
  for (let i = streamDays.length - 1; i >= 0; i--) {
268
364
  const day = streamDays[i]!
269
365
  for (const event of day.events) {
270
- const match = matchStreamEvent(day, event, matcher, options.full)
366
+ const match = matchStreamEvent(day, event, matcher, options.full, options.streamDocuments)
271
367
  if (match === null) continue
272
368
  if (!push(match)) return { matches, truncatedAt: truncatedAt! }
273
369
  }
@@ -299,6 +395,9 @@ export function searchAllRanked(
299
395
  maxResults: number
300
396
  references?: Reference[]
301
397
  tokenMatchMode?: 'substring' | 'ascii-boundary'
398
+ topicDocuments?: Map<string, string>
399
+ topicProvenance?: Map<string, ProvenanceChild[]>
400
+ streamDocuments?: Map<string, string>
302
401
  },
303
402
  ): MemorySearchResult {
304
403
  const tokenMatchers = tokens.map((t) => buildTokenMatcher(t, options.tokenMatchMode ?? 'substring'))
@@ -315,9 +414,19 @@ export function searchAllRanked(
315
414
  let order = 0
316
415
 
317
416
  for (const shard of shards) {
318
- const match = matchShard(shard, anyToken, options.full)
417
+ const match = matchShard(
418
+ shard,
419
+ anyToken,
420
+ options.full,
421
+ options.topicDocuments?.get(shard.slug),
422
+ options.topicProvenance?.get(shard.slug),
423
+ )
319
424
  if (match === null) continue
320
- scored.push({ match, score: scoreOf(shardSearchText(shard)), order: order++ })
425
+ scored.push({
426
+ match,
427
+ score: scoreOf(options.topicDocuments?.get(shard.slug) ?? shardSearchText(shard)),
428
+ order: order++,
429
+ })
321
430
  }
322
431
 
323
432
  for (const reference of options.references ?? []) {
@@ -329,9 +438,16 @@ export function searchAllRanked(
329
438
  for (let i = streamDays.length - 1; i >= 0; i--) {
330
439
  const day = streamDays[i]!
331
440
  for (const event of day.events) {
332
- const match = matchStreamEvent(day, event, anyToken, options.full)
441
+ const match = matchStreamEvent(day, event, anyToken, options.full, options.streamDocuments)
333
442
  if (match === null) continue
334
- scored.push({ match, score: scoreOf(eventSearchText(event)), order: order++ })
443
+ const citation = event.type === 'fragment' ? `streams/${day.date}#${event.id}` : undefined
444
+ scored.push({
445
+ match,
446
+ score: scoreOf(
447
+ (citation === undefined ? undefined : options.streamDocuments?.get(citation)) ?? eventSearchText(event),
448
+ ),
449
+ order: order++,
450
+ })
335
451
  }
336
452
  }
337
453
 
@@ -378,7 +494,13 @@ function referenceSearchText(reference: Reference): string {
378
494
  return [reference.slug, reference.frontmatter.title, ...reference.frontmatter.tags, reference.body].join('\n')
379
495
  }
380
496
 
381
- function matchShard(shard: TopicShard, matcher: Matcher, full: boolean): TopicMatch | null {
497
+ function matchShard(
498
+ shard: TopicShard,
499
+ matcher: Matcher,
500
+ full: boolean,
501
+ searchText = shardSearchText(shard),
502
+ provenance?: ProvenanceChild[],
503
+ ): TopicMatch | null {
382
504
  const bodyLines = splitBodyLines(shard.body)
383
505
  const firstBodyLineIndex = bodyLines.findIndex((line) => matcher(line))
384
506
 
@@ -386,7 +508,8 @@ function matchShard(shard: TopicShard, matcher: Matcher, full: boolean): TopicMa
386
508
  matcher(shard.slug) ||
387
509
  matcher(shard.frontmatter.heading) ||
388
510
  (shard.frontmatter.tags?.some((tag) => matcher(tag)) ?? false) ||
389
- firstBodyLineIndex !== -1
511
+ firstBodyLineIndex !== -1 ||
512
+ matcher(searchText)
390
513
  if (!matched) return null
391
514
 
392
515
  const match: TopicMatch = {
@@ -398,6 +521,7 @@ function matchShard(shard: TopicShard, matcher: Matcher, full: boolean): TopicMa
398
521
  firstBodyLineIndex === -1 ? fallbackExcerpt(shard, matcher) : excerptForLine(bodyLines, firstBodyLineIndex),
399
522
  }
400
523
  if (full) match.fullBody = shard.body
524
+ if (provenance !== undefined && provenance.length > 0) match.provenance = provenance
401
525
  return match
402
526
  }
403
527
 
@@ -439,9 +563,12 @@ function matchStreamEvent(
439
563
  event: StreamEvent,
440
564
  matcher: Matcher,
441
565
  full: boolean,
566
+ streamDocuments?: Map<string, string>,
442
567
  ): StreamMatch | null {
443
568
  if (event.type === 'watermark') return null
444
- if (event.type === 'fragment') return matchFragmentEvent(day, event, matcher, full)
569
+ if (event.type === 'fragment') {
570
+ return matchFragmentEvent(day, event, matcher, full, streamDocuments?.get(`streams/${day.date}#${event.id}`))
571
+ }
445
572
  return matchLegacyProseEvent(day, event, matcher, full)
446
573
  }
447
574
 
@@ -450,10 +577,11 @@ function matchFragmentEvent(
450
577
  event: FragmentEvent,
451
578
  matcher: Matcher,
452
579
  full: boolean,
580
+ searchText = eventSearchText(event),
453
581
  ): StreamMatch | null {
454
582
  const bodyLines = splitBodyLines(event.body)
455
583
  const firstBodyLineIndex = bodyLines.findIndex((line) => matcher(line))
456
- const matched = matcher(event.topic) || firstBodyLineIndex !== -1
584
+ const matched = matcher(event.topic) || firstBodyLineIndex !== -1 || matcher(searchText)
457
585
  if (!matched) return null
458
586
 
459
587
  const match: StreamMatch = {
@@ -465,7 +593,8 @@ function matchFragmentEvent(
465
593
  excerpt: firstBodyLineIndex === -1 ? event.topic : excerptForLine(bodyLines, firstBodyLineIndex),
466
594
  when: event.ts,
467
595
  }
468
- if (event.who !== undefined) match.who = event.who
596
+ const who = sanitizeProvenanceName(event.who)
597
+ if (who !== undefined) match.who = who
469
598
  if (event.where !== undefined) match.where = event.where
470
599
  if (full) match.fullBody = event.body
471
600
  return match
@@ -530,32 +659,6 @@ function parseDateParam(name: string, value: string): Date | string {
530
659
  return date
531
660
  }
532
661
 
533
- // Keep only fragments whose `where` names this room, by raw `chat` id (canonical)
534
- // or `chatName` (convenience, case-insensitive — the agent often knows only
535
- // "#incidents"). The `#` prefix is stripped so "#incidents" and "incidents"
536
- // both match. Non-fragment events (watermarks, legacy prose) have no `where` and
537
- // are dropped under a room filter; days that end up empty are removed.
538
- function filterStreamDaysByRoom(days: UndreamedStreamDay[], room: string): UndreamedStreamDay[] {
539
- const normalizedKey = normalizeRoomKey(room)
540
- return days.flatMap((day) => {
541
- const events = day.events.filter(
542
- (event) => event.type === 'fragment' && fragmentMatchesRoom(event, room, normalizedKey),
543
- )
544
- return events.length === 0 ? [] : [{ ...day, events }]
545
- })
546
- }
547
-
548
- function fragmentMatchesRoom(event: FragmentEvent, rawRoom: string, normalizedKey: string): boolean {
549
- const where = event.where
550
- if (where === undefined) return false
551
- if (where.chat === rawRoom) return true
552
- return where.chatName !== undefined && normalizeRoomKey(where.chatName) === normalizedKey
553
- }
554
-
555
- function normalizeRoomKey(value: string): string {
556
- return value.replace(/^#/, '').toLocaleLowerCase()
557
- }
558
-
559
662
  function referenceCandidateAllowed(reference: Reference, filter: ReferenceDateFilter): boolean {
560
663
  if (reference.frontmatter.demoted) return false
561
664
  const created = new Date(reference.frontmatter.created)
@@ -42,6 +42,8 @@ export const fragmentProvenanceSchema = z
42
42
  chat: z.string(),
43
43
  chatName: z.string().optional(),
44
44
  thread: z.string().nullable().optional(),
45
+ parentChat: z.string().optional(),
46
+ parentChatName: z.string().optional(),
45
47
  })
46
48
  .passthrough()
47
49