switchroom 0.19.18 → 0.19.22

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 (76) hide show
  1. package/dist/agent-scheduler/index.js +2 -1
  2. package/dist/auth-broker/index.js +56 -1
  3. package/dist/cli/drive-write-pretool.mjs +48 -5
  4. package/dist/cli/ms-365-write-pretool.mjs +40 -2
  5. package/dist/cli/notion-write-pretool.mjs +2 -1
  6. package/dist/cli/switchroom.js +5242 -2239
  7. package/dist/host-control/main.js +12241 -11375
  8. package/dist/vault/approvals/kernel-server.js +113 -7
  9. package/dist/vault/broker/server.js +259 -76
  10. package/package.json +6 -3
  11. package/profiles/_base/start.sh.hbs +61 -1
  12. package/skills/switchroom-release/SKILL.md +103 -20
  13. package/telegram-plugin/bridge/bridge.ts +14 -0
  14. package/telegram-plugin/card-format.ts +92 -3
  15. package/telegram-plugin/dist/bridge/bridge.js +13 -0
  16. package/telegram-plugin/dist/gateway/gateway.js +2356 -1159
  17. package/telegram-plugin/dist/server.js +13 -0
  18. package/telegram-plugin/edit-flood-fuse.ts +477 -0
  19. package/telegram-plugin/format.ts +19 -7
  20. package/telegram-plugin/gateway/always-allow-persist-queue.ts +97 -11
  21. package/telegram-plugin/gateway/boot-sweep-gate.ts +164 -0
  22. package/telegram-plugin/gateway/callback-query-handlers.ts +454 -81
  23. package/telegram-plugin/gateway/gateway.ts +66 -56
  24. package/telegram-plugin/gateway/inbound-interceptors.ts +27 -4
  25. package/telegram-plugin/gateway/missed-approvals-store.ts +66 -17
  26. package/telegram-plugin/gateway/narrative-lane.ts +49 -3
  27. package/telegram-plugin/gateway/pending-card-store.ts +46 -16
  28. package/telegram-plugin/gateway/scoped-grant-store.ts +39 -14
  29. package/telegram-plugin/gateway/status-pin-api.ts +145 -0
  30. package/telegram-plugin/gateway/store-file.ts +244 -0
  31. package/telegram-plugin/hooks/subagent-tracker-posttool.mjs +325 -45
  32. package/telegram-plugin/hooks/tool-label-pretool.mjs +88 -2
  33. package/telegram-plugin/retry-api-call.ts +15 -2
  34. package/telegram-plugin/send-gate.ts +1 -1
  35. package/telegram-plugin/status-no-truncate.ts +64 -1
  36. package/telegram-plugin/status-pin-driver.ts +50 -27
  37. package/telegram-plugin/status-pin.ts +43 -5
  38. package/telegram-plugin/tests/activity-card-send-gate.test.ts +275 -0
  39. package/telegram-plugin/tests/activity-card-wiring.test.ts +16 -7
  40. package/telegram-plugin/tests/boot-pin-sweep-wiring.test.ts +101 -0
  41. package/telegram-plugin/tests/boot-sweep-gate.test.ts +293 -0
  42. package/telegram-plugin/tests/boot-version-string.test.ts +0 -0
  43. package/telegram-plugin/tests/bridge-tool-parity.test.ts +95 -0
  44. package/telegram-plugin/tests/edit-flood-fuse.test.ts +431 -0
  45. package/telegram-plugin/tests/pinned-card-collapse.test.ts +356 -0
  46. package/telegram-plugin/tests/status-pin-api.test.ts +178 -0
  47. package/telegram-plugin/tests/status-pin-boot-recovery.test.ts +94 -11
  48. package/telegram-plugin/tests/status-pin.test.ts +106 -5
  49. package/telegram-plugin/tests/store-atomic-write.test.ts +411 -0
  50. package/telegram-plugin/tests/subagent-tracker-hooks.test.ts +631 -1
  51. package/telegram-plugin/tests/tool-activity-summary.test.ts +28 -12
  52. package/telegram-plugin/tests/tool-label-pretool.test.ts +94 -0
  53. package/telegram-plugin/tests/vault-approval-posture.test.ts +6 -1
  54. package/telegram-plugin/tests/vault-passphrase-retry.test.ts +666 -0
  55. package/telegram-plugin/tests/vault-request-access-unlock-resume.test.ts +42 -21
  56. package/telegram-plugin/tests/worker-feed-coalesce.test.ts +233 -1
  57. package/telegram-plugin/tests/worker-feed-repeat-steps.test.ts +147 -0
  58. package/telegram-plugin/tool-activity-summary.ts +85 -13
  59. package/telegram-plugin/worker-activity-feed.ts +56 -2
  60. package/vendor/hindsight-memory/scripts/drain_pending.py +847 -67
  61. package/vendor/hindsight-memory/scripts/lib/client.py +124 -0
  62. package/vendor/hindsight-memory/scripts/lib/pending.py +944 -33
  63. package/vendor/hindsight-memory/scripts/lib/retain_split.py +460 -0
  64. package/vendor/hindsight-memory/scripts/recall.py +74 -5
  65. package/vendor/hindsight-memory/scripts/session_start.py +48 -0
  66. package/vendor/hindsight-memory/scripts/tests/test_client_document_exists.py +470 -0
  67. package/vendor/hindsight-memory/scripts/tests/test_pending_drops.py +2275 -0
  68. package/vendor/hindsight-memory/scripts/tests/test_pending_failure_class.py +105 -0
  69. package/vendor/hindsight-memory/scripts/tests/test_pending_wedge.py +300 -0
  70. package/vendor/hindsight-memory/scripts/tests/test_recall_degraded_notice.py +365 -0
  71. package/vendor/hindsight-memory/scripts/tests/test_recall_envelope_strip_telemetry.py +12 -4
  72. package/vendor/hindsight-memory/scripts/tests/test_recall_transcript_fallback.py +27 -2
  73. package/vendor/hindsight-memory/scripts/tests/test_retain_split.py +438 -0
  74. package/vendor/hindsight-memory/scripts/tests/test_session_start_version_skew.py +204 -0
  75. package/vendor/hindsight-memory/tests/test_drain_pending.py +130 -8
  76. package/vendor/hindsight-memory/tests/test_pending.py +32 -7
@@ -89,6 +89,7 @@ import {
89
89
  WORKER_HISTORY_MAX,
90
90
  STATUS_LINE_MAX,
91
91
  NESTED_PREFIX,
92
+ WORKER_STEP_INDENT,
92
93
  } from './status-no-truncate.js'
93
94
  import { escapeMarkdown, stripMarkdown, truncate, stackCardLines } from './card-format.js'
94
95
  import { isTelegramSurfaceTool } from './tool-names.js'
@@ -297,6 +298,11 @@ function escapeStepLine(raw: string): string {
297
298
  * `allDone` — when true ALL steps render done (✓ struck italic); when false the
298
299
  * newest renders in-progress (→ bold)
299
300
  * `liveSuffix` — appended INSIDE the newest in-progress line (heartbeat tick)
301
+ * `indent` — literal prefix put on EVERY emitted line (incl. the
302
+ * `+N earlier…` header), OUTSIDE the markdown spans so it never
303
+ * lands inside an emphasis run. Default `''` — byte-identical
304
+ * output for callers that don't indent. The combined worker card
305
+ * passes `WORKER_STEP_INDENT` to nest steps under their worker.
300
306
  */
301
307
  export function renderStepFeed(
302
308
  out: string[],
@@ -304,14 +310,19 @@ export function renderStepFeed(
304
310
  allDone: boolean,
305
311
  liveSuffix = '',
306
312
  window: number = STATUS_ROLLING_LINES,
313
+ indent = '',
307
314
  ): void {
308
315
  if (steps.length === 0) return
309
316
  const shown = steps.slice(-Math.max(1, window))
310
317
  const hidden = steps.length - shown.length
311
- if (hidden > 0) out.push(`_✓ +${hidden} earlier…_`)
318
+ if (hidden > 0) out.push(`${indent}_✓ +${hidden} earlier…_`)
312
319
  const lastIdx = shown.length - 1
313
320
  shown.forEach((s, i) => {
314
- out.push(!allDone && i === lastIdx ? `**→ ${s}${liveSuffix}**` : `~~_✓ ${s}_~~`)
321
+ out.push(
322
+ !allDone && i === lastIdx
323
+ ? `${indent}**→ ${s}${liveSuffix}**`
324
+ : `${indent}~~_✓ ${s}_~~`,
325
+ )
315
326
  })
316
327
  }
317
328
 
@@ -439,7 +450,10 @@ export function renderStatusCard(opts: StatusCardOpts): string | null {
439
450
  // Stack lines with GFM hard breaks (` \n`) so the card's styled prose lines
440
451
  // don't collapse onto one visual line in the rich-message renderer — see
441
452
  // stackCardLines. This is what makes a card render identically to a reply.
442
- const joined = stackCardLines(out)
453
+ // collapseSafe (#3666): the agent status card and the single-worker card are
454
+ // both PINNED, and Telegram's pinned bar shows them collapsed to one line
455
+ // with the newlines dropped — the separator keeps that preview legible.
456
+ const joined = stackCardLines(out, { collapseSafe: true })
443
457
  if (joined.length <= STATUS_CARD_CHAR_BUDGET) return joined
444
458
  return fitCardToBudget(opts, headerLines)
445
459
  }
@@ -492,7 +506,8 @@ function fitCardToBudget(opts: StatusCardOpts, headerLines: string[]): string {
492
506
  const lastIdx = shown.length - 1
493
507
  shown.forEach((esc, i) => lines.push(buildBullet(esc, i === lastIdx)))
494
508
  lines.push(...footerLines)
495
- const candidate = stackCardLines(lines)
509
+ // collapseSafe: same pinned surface as renderStatusCard (#3666).
510
+ const candidate = stackCardLines(lines, { collapseSafe: true })
496
511
  if (candidate.length <= STATUS_CARD_CHAR_BUDGET) return candidate
497
512
  }
498
513
 
@@ -518,7 +533,7 @@ function fitCardToBudget(opts: StatusCardOpts, headerLines: string[]): string {
518
533
  if (parentMarker != null) lines.push(parentMarker)
519
534
  lines.push(newestLine)
520
535
  lines.push(...footerLines)
521
- return stackCardLines(lines)
536
+ return stackCardLines(lines, { collapseSafe: true })
522
537
  }
523
538
 
524
539
  /**
@@ -729,18 +744,54 @@ export function combinedHistoryDepth(w: number): number {
729
744
  /** Alias for readability at the single-worker call site — same curve. */
730
745
  export const workerHistoryDepth = combinedHistoryDepth
731
746
 
747
+ /**
748
+ * The combined card's first line: a SELF-CONTAINED glance at the whole swarm
749
+ * (#3666) — `🛠 Workers · 3 running · oldest 38m16s · 196 tools · 1.2M tok`.
750
+ *
751
+ * Two jobs, both load-bearing:
752
+ *
753
+ * 1. In the feed it answers "what is my fleet doing" without reading the
754
+ * rows — the aggregate the per-row lines never showed.
755
+ * 2. In Telegram's pinned bar, where the card collapses to ONE line, it is
756
+ * the part most likely to survive truncation, so the preview leads with
757
+ * real information instead of a four-word chrome stub that ran into
758
+ * row 1's ordinal (`… 3 running1. Fix issue …`).
759
+ *
760
+ * It always ends in a UNIT WORD (`running` / `oldest <elapsed>` / `tools` /
761
+ * `tok`), never a bare number, so the first collapse seam reads as
762
+ * `… · 196 tools 1. Fix issue …` — a list starting after a unit — rather than
763
+ * two numbers colliding. Aggregates cover ALL rows (including any spilled to
764
+ * `+M more working…`), matching the `N running` count beside them.
765
+ */
766
+ function glanceLine(rows: CombinedWorkerRow[]): string {
767
+ const oldestMs = rows.reduce((m, r) => Math.max(m, r.elapsedMs), 0)
768
+ const tools = rows.reduce((n, r) => n + r.toolCount, 0)
769
+ const tok = rows.reduce((n, r) => n + (r.totalTokens ?? 0), 0)
770
+ const toolWord = tools === 1 ? 'tool' : 'tools'
771
+ return (
772
+ `🛠 **Workers** · _${rows.length} running · oldest ${formatFeedElapsed(oldestMs)}` +
773
+ ` · ${tools} ${toolWord}${tokenSegment(tok)}_`
774
+ )
775
+ }
776
+
732
777
  /**
733
778
  * Render N≥1 live workers into ONE combined feed body (ready Telegram
734
779
  * markdown; callers send verbatim — do NOT re-escape). Layout:
735
780
  *
736
- * 🛠 **Workers** · _N running_
781
+ * 🛠 **Workers** · _N running · oldest {elapsed} · {n} tools · {t} tok_
737
782
  * **1. {desc1}** _· {elapsed} · {n} tools_
738
- * ~~_✓ {earlier step}_~~
739
- * **→ {newest step}**
783
+ * ~~_✓ {earlier step}_~~
784
+ * **→ {newest step}**
740
785
  * **2. {desc2}** _· {elapsed} · {n} tools_
741
- * **→ {newest step}**
786
+ * **→ {newest step}**
742
787
  * _+M more working…_
743
788
  *
789
+ * INDENT: every step line carries a leading `WORKER_STEP_INDENT` (a U+2800 run
790
+ * — neither ASCII spaces nor U+00A0; Telegram left-trims BOTH, see the
791
+ * constant's doc comment) so the steps nest under their worker header and the
792
+ * per-worker blocks are scannable. Header and chrome lines stay at the left
793
+ * margin.
794
+ *
744
795
  * NUMBERING (#3298): when the card tracks 2+ rows AND a row carries `ordinal`,
745
796
  * its header gets a stable `{ordinal}. ` prefix. Ordinals are assigned by the
746
797
  * caller at dispatch and kept for the card's life — after an earlier worker
@@ -755,6 +806,15 @@ export const workerHistoryDepth = combinedHistoryDepth
755
806
  * card staying bounded regardless of fan-out. When a worker has no history yet
756
807
  * it falls back to a single `→ starting…`/currentStep line.
757
808
  *
809
+ * PINNED-BAR READABILITY (#3666): line 1 is a self-contained glance
810
+ * (`glanceLine`) and the stack is joined with `collapseSafe`, because this card
811
+ * is pinned and Telegram's pinned bar renders it collapsed onto ONE line with
812
+ * the newlines dropped and NOTHING substituted. Both halves are needed: the
813
+ * glance so the preview leads with information, the per-line separator so every
814
+ * later seam (`opus 5` → `✓ …` → `→ …`) degrades to a word gap instead of a
815
+ * mashed glyph. There is no Bot API for a separate pin-bar preview — the client
816
+ * derives it from the message text — so the text is the only lever.
817
+ *
758
818
  * Pure. Rows are rendered in the order supplied (the manager passes them
759
819
  * dispatch-order, oldest first). `maxRows` caps the visible rows; the hidden
760
820
  * remainder collapses to a single `+M more working…` line. A total-budget
@@ -802,13 +862,13 @@ export function renderCombinedWorkerFeed(
802
862
  // Per-worker depth follows Ken's deterministic curve max(3, 7−w) (#3349):
803
863
  // the curve drives DEPTH; the total-line budget below drives ROW COUNT.
804
864
  const depth = combinedHistoryDepth(shown.length)
805
- const chrome: string[] = [`🛠 **Workers** · _${rows.length} running_`]
865
+ const chrome: string[] = [glanceLine(rows)]
806
866
  const bodyOut: string[] = []
807
867
  for (const r of shown) {
808
868
  bodyOut.push(rowHeader(r))
809
869
  const hist = rowHistory(r)
810
870
  if (hist.length === 0) {
811
- bodyOut.push('→ _starting…_')
871
+ bodyOut.push(`${WORKER_STEP_INDENT}→ _starting…_`)
812
872
  continue
813
873
  }
814
874
  // Paint the last-K history lines with the SAME `✓`/`→` idiom as the
@@ -816,12 +876,24 @@ export function renderCombinedWorkerFeed(
816
876
  // pipeline (escapeStepLine), then renderStepFeed strikes the prior steps
817
877
  // and bolds the newest in-progress step. The window equals the depth so a
818
878
  // per-worker `+N earlier…` marker never appears inside the combined feed.
879
+ //
880
+ // WORKER_STEP_INDENT nests every step line one level under its worker
881
+ // header, so the boundary between two workers is visible at a glance on a
882
+ // phone (the header lines stay at the left margin, their steps sit in).
883
+ // It is a U+2800 run. NOT ASCII spaces and NOT U+00A0: Telegram's
884
+ // server-side parser left-trims a leading Unicode-whitespace run, so both
885
+ // render flat (#3662 shipped U+00A0 and was inert). U+2800 is category So,
886
+ // not Zs. See the constant's doc comment for the live evidence.
819
887
  const esc = hist.slice(-depth).map(escapeStepLine)
820
- renderStepFeed(bodyOut, esc, false, '', depth)
888
+ renderStepFeed(bodyOut, esc, false, '', depth, WORKER_STEP_INDENT)
821
889
  }
822
890
  const out = [...chrome, ...bodyOut]
823
891
  if (hidden > 0) out.push(`_+${hidden} more working…_`)
824
- return { body: stackCardLines(out), bodyLines: bodyOut.length }
892
+ // collapseSafe (#3666): this card is PINNED, and Telegram's pinned bar
893
+ // renders it collapsed to one line with the newlines dropped and nothing
894
+ // substituted. Without the separator the glance line runs straight into
895
+ // row 1's ordinal and every step glyph mashes into the previous line.
896
+ return { body: stackCardLines(out, { collapseSafe: true }), bodyLines: bodyOut.length }
825
897
  }
826
898
 
827
899
  // Cap to maxRows first, then shrink the visible set while EITHER the total
@@ -53,6 +53,7 @@ import {
53
53
  cleanWorkerResultParagraph,
54
54
  stripMarkdown,
55
55
  truncate,
56
+ COLLAPSE_SAFE_SEPARATOR,
56
57
  } from './card-format.js'
57
58
  import { WORKER_HISTORY_MAX } from './status-no-truncate.js'
58
59
  import {
@@ -142,6 +143,25 @@ export interface BotApiForWorkerFeed {
142
143
  /** Dispatch-time task description cap for the worker header. */
143
144
  const DESC_MAX = 80
144
145
 
146
+ /**
147
+ * Repeat marker appended to a narrative step that fired more than once in a
148
+ * row (`Running a command ·×3`). Rendered inline on the step line so a worker
149
+ * whose every step carries the SAME label still visibly advances instead of
150
+ * freezing the card. `·` is already the card's separator glyph.
151
+ */
152
+ const REPEAT_SUFFIX_RE = / ·×(\d+)$/
153
+
154
+ /** The step text without its `·×N` marker (identity for dedup comparisons). */
155
+ export function stripRepeatSuffix(line: string): string {
156
+ return line.replace(REPEAT_SUFFIX_RE, '')
157
+ }
158
+
159
+ /** How many times a step line has fired: 1 when it carries no marker. */
160
+ export function repeatCountOf(line: string): number {
161
+ const m = REPEAT_SUFFIX_RE.exec(line)
162
+ return m == null ? 1 : Number(m[1])
163
+ }
164
+
145
165
  /**
146
166
  * Thin adapter over the unified `renderStatusCard` primitive (emoji 🛠, label
147
167
  * 'Worker'): builds the header, passes raw narrative steps (the primitive runs
@@ -226,7 +246,10 @@ export function renderWorkerActivity(v: WorkerActivityView, liveSuffix = ''): st
226
246
  // Header-only running render → append the starting placeholder with a GFM
227
247
  // hard break (` \n`) so it stacks under the header instead of collapsing
228
248
  // onto the header line in the rich-message renderer (matches stackCardLines).
229
- return `${card} \n_starting…_`
249
+ // The collapse separator is carried here too (#3666) — this card is pinned,
250
+ // and a hand-rolled seam would be the one boundary that still mashed in
251
+ // Telegram's pinned bar.
252
+ return `${card}${COLLAPSE_SAFE_SEPARATOR} \n_starting…_`
230
253
  }
231
254
  return card
232
255
  }
@@ -428,6 +451,13 @@ interface WorkerRow {
428
451
  * live render so the feed reads like the main agent's answer.
429
452
  */
430
453
  narrative: string[]
454
+ /**
455
+ * `view.toolCount` observed when the newest narrative step was last recorded
456
+ * or counted. The repeat counter (`·×N`) increments only when toolCount has
457
+ * MOVED — the watcher re-emits an unchanged view every tick, so counting
458
+ * label-equality alone would inflate the number with no work behind it.
459
+ */
460
+ lastNarrativeToolCount: number | null
431
461
  /** Last view for this worker (drives the heartbeat re-render + combined row). */
432
462
  lastView: WorkerActivityView | null
433
463
  /** Latest state observed for this worker; excluded from the running set once
@@ -794,10 +824,33 @@ export function createWorkerActivityFeed(opts: WorkerActivityFeedOpts): WorkerAc
794
824
  function accumulateNarrative(row: WorkerRow, view: WorkerActivityView): void {
795
825
  const line = view.latestSummary.trim()
796
826
  if (line.length === 0) return
827
+
828
+ // A REPEAT of the current step is counted, not dropped. A worker whose
829
+ // steps all render the same label (e.g. Bash calls with no description →
830
+ // "Running a command") used to hit the dedup below on every tool call and
831
+ // the card held ONE frozen line for the whole job — indistinguishable from
832
+ // a wedged worker. `·×N` makes the repetition visible and, because a new
833
+ // count resets stepStartedAtMs, the climbing `· Ns` step timer restarts
834
+ // with each real call.
835
+ //
836
+ // Gated on `view.toolCount`, NOT on the label: the watcher re-emits an
837
+ // UNCHANGED view on every tick, so counting label-equality alone would
838
+ // inflate the number with no work behind it. One increment per observed
839
+ // tool call, deterministically.
840
+ const last = row.narrative.length > 0 ? row.narrative[row.narrative.length - 1] : null
841
+ if (last != null && stripRepeatSuffix(last) === line) {
842
+ if (view.toolCount === row.lastNarrativeToolCount) return
843
+ row.lastNarrativeToolCount = view.toolCount
844
+ row.narrative[row.narrative.length - 1] = `${line} ·×${repeatCountOf(last) + 1}`
845
+ row.stepStartedAtMs = nowFn()
846
+ return
847
+ }
848
+
797
849
  // Dedup within the whole rolling window (the watcher re-emits the same
798
850
  // narrative across ticks, and a preamble + its tool label can repeat
799
851
  // non-adjacently — the A,B,A duplication observed on live cards).
800
- if (row.narrative.includes(line)) return
852
+ if (row.narrative.some((l) => stripRepeatSuffix(l) === line)) return
853
+ row.lastNarrativeToolCount = view.toolCount
801
854
  row.narrative.push(line)
802
855
  // The `→` current-step line just CHANGED — reset the per-step timer.
803
856
  row.stepStartedAtMs = nowFn()
@@ -1545,6 +1598,7 @@ export function createWorkerActivityFeed(opts: WorkerActivityFeedOpts): WorkerAc
1545
1598
  agentId,
1546
1599
  ordinal: ++g.workerOrdinalCounter,
1547
1600
  narrative: [],
1601
+ lastNarrativeToolCount: null,
1548
1602
  lastView: null,
1549
1603
  state: 'running',
1550
1604
  finished: false,