samagotchi 0.125.0 → 0.126.0

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 (50) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +25 -1
  3. data/docs/guardrails.md +22 -0
  4. data/docs/internals/participants.md +70 -2
  5. data/docs/internals/web-push.md +10 -6
  6. data/docs/sessions.md +3 -0
  7. data/lib/samagotchi/bridge.rb +4 -3
  8. data/lib/samagotchi/bridge_client.rb +8 -1
  9. data/lib/samagotchi/bundles/known-names/hooks/known_names.rb +25 -4
  10. data/lib/samagotchi/bundles/known-names/manifest.yml +2 -2
  11. data/lib/samagotchi/bundles/system/manifest.yml +1 -1
  12. data/lib/samagotchi/child_reports.rb +3 -0
  13. data/lib/samagotchi/children_status.rb +6 -1
  14. data/lib/samagotchi/engine.rb +43 -7
  15. data/lib/samagotchi/guardrail_wiring.rb +10 -1
  16. data/lib/samagotchi/llm/chat_loop.rb +1 -0
  17. data/lib/samagotchi/session.rb +40 -3
  18. data/lib/samagotchi/session_inbox.rb +14 -8
  19. data/lib/samagotchi/session_manager.rb +11 -4
  20. data/lib/samagotchi/session_participants.rb +10 -2
  21. data/lib/samagotchi/verdict.rb +20 -0
  22. data/lib/samagotchi/verify_anchor.rb +28 -0
  23. data/lib/samagotchi/verify_fresh.rb +275 -0
  24. data/lib/samagotchi/verify_fresh_origin.rb +71 -0
  25. data/lib/samagotchi/version.rb +1 -1
  26. data/lib/samagotchi/web/app.rb +3 -0
  27. data/lib/samagotchi/web/participant_access.rb +15 -4
  28. data/lib/samagotchi/web/participant_review.rb +7 -1
  29. data/lib/samagotchi/web/participant_view.rb +46 -5
  30. data/lib/samagotchi/web/public/annotate_bar.js +9 -1
  31. data/lib/samagotchi/web/public/app.js +69 -4
  32. data/lib/samagotchi/web/public/data.js +2 -1
  33. data/lib/samagotchi/web/public/index.html +1 -0
  34. data/lib/samagotchi/web/public/notify.js +13 -2
  35. data/lib/samagotchi/web/public/participant.html +2 -1
  36. data/lib/samagotchi/web/public/participant.js +69 -7
  37. data/lib/samagotchi/web/public/send_turn.js +6 -1
  38. data/lib/samagotchi/web/public/session_state.js +5 -0
  39. data/lib/samagotchi/web/public/stream_handlers.js +3 -0
  40. data/lib/samagotchi/web/public/verdict_card.js +65 -0
  41. data/lib/samagotchi/web/public/verdict_fresh.js +120 -0
  42. data/lib/samagotchi/web/public/verify_card.css +17 -0
  43. data/lib/samagotchi/web/push/attention.rb +25 -5
  44. data/lib/samagotchi/web/push/notifier.rb +4 -3
  45. data/lib/samagotchi/web/session_control.rb +2 -0
  46. data/lib/samagotchi/web/session_summary.rb +25 -1
  47. data/lib/samagotchi/web/verify_fresh_routes.rb +110 -0
  48. data/lib/samagotchi/worker.rb +68 -9
  49. data/lib/samagotchi/worker_inbound.rb +4 -4
  50. metadata +9 -1
checksums.yaml CHANGED
@@ -1,7 +1,7 @@
1
1
  ---
2
2
  SHA256:
3
- metadata.gz: 8d36716e229ef5ccac918c840a3b55d878257811aeffdf594d2588c3bf067533
4
- data.tar.gz: fb632ff225b0a9f5dfbbc869ce8df2842df811b6858caa3d714362879a3e4287
3
+ metadata.gz: 2704d1d3f6cb8fa560e7f903ac58ac12d468fcd006b2a8782dcea02aede223a8
4
+ data.tar.gz: 49f91663fa29e4f77b41dae067efe8bba85917eef2682f1911af7ed7225db8fd
5
5
  SHA512:
6
- metadata.gz: d16837760eae98a5f26bef1349f70d6fecf80b2400858d2fab12d0a39b11cb5ed125901b21e99bede021585cd9ad60c44adb6a3dad0416a7c4302ebaed459950
7
- data.tar.gz: 8210416816fda4da83f18801d6e2b991335c41054cfd5459361c9b478d5c633a02ac58855d9f77520d05269c2f64e004cd600c7e56b36ac12d71efe612965430
6
+ metadata.gz: 0f2ed298f6d733e3f90e926b2124348c40fc119c934106d9661aa3fdb0cfd060535263516d31789ee5be3a043abe3a4c3c844754323d6bdf9e44e527587739b4
7
+ data.tar.gz: e3d1196a4881d05ae6fcc9f94700cc4e180a024b9d725de02baa42de3d46ae1b78146a8493aa5b42c707919476f410d23e059e9df24e3d9c1fa955362921afe3
data/CHANGELOG.md CHANGED
@@ -7,6 +7,29 @@ and commands may change between minor versions. How releases are made:
7
7
  [docs/releasing.md](docs/releasing.md).
8
8
  ## [Unreleased]
9
9
 
10
+ ## [0.126.0] - 2026-10-27
11
+
12
+ ### Added
13
+
14
+ - chi web places each completed Verify verdict on its source claim, links to the checking turn, and sends a phone
15
+ notification with the verdict and a short reason (including for short turns).
16
+ - **Verify fresh** on an unclear verdict's card (owner or participant): a new agent checks the claim in the session's
17
+ folder with none of its context; its verdict lands on the same card. A participant's runs under their own access
18
+ and spend cap.
19
+
20
+ ### Changed
21
+
22
+ - A pair participant's page header reads "<name> · pairing" instead of "<name>'s review", and its tools overview's
23
+ disclosure triangle matches the text's size.
24
+
25
+ ### Fixed
26
+
27
+ - The known-names hook lets through a protected name's CamelCase form (`CoffeeAssistant` or `coffeeAssistant` for a
28
+ repo named `coffee_assistant`) instead of rejecting it as a misspelling; `coffee-assistant` and `coffeeassistant`
29
+ are still caught.
30
+ - The known-names hook ignores separators at a word's start or end, so `grep coffee_assistant-` or
31
+ `.coffee_assistant` is no longer rejected as a misspelling; `mode: correct` keeps those separators.
32
+
10
33
  ## [0.125.0] - 2026-10-27
11
34
 
12
35
  ### Added
@@ -4740,7 +4763,8 @@ and long-lived sessions.
4740
4763
  - A macOS desktop helper (`chi desktop install`): a "Send to chi" Service and a
4741
4764
  hotkey panel that send selected text or the clipboard to your sessions.
4742
4765
 
4743
- [Unreleased]: https://github.com/dm1try/samagotchi/compare/v0.125.0...HEAD
4766
+ [Unreleased]: https://github.com/dm1try/samagotchi/compare/v0.126.0...HEAD
4767
+ [0.126.0]: https://github.com/dm1try/samagotchi/compare/v0.125.0...v0.126.0
4744
4768
  [0.125.0]: https://github.com/dm1try/samagotchi/compare/v0.124.0...v0.125.0
4745
4769
  [0.124.0]: https://github.com/dm1try/samagotchi/compare/v0.123.0...v0.124.0
4746
4770
  [0.123.0]: https://github.com/dm1try/samagotchi/compare/v0.122.0...v0.123.0
data/docs/guardrails.md CHANGED
@@ -452,6 +452,28 @@ can send out in a URL), plugin and MCP tools, `memory_read` and `context_read`,
452
452
  does that the command checks don't read (a script file's body, an interpreter's
453
453
  `-e` code). Share a working session only with people you would let type in it.
454
454
 
455
+ ## A Verify fresh child's boundary
456
+
457
+ A [Verify fresh](internals/participants.md#what-a-participant-sees) child (a
458
+ delegate child chi web starts for an unclear verdict) runs a participant's
459
+ check under **the boundary that participant has in the parent**. A verdict a
460
+ participant recorded is always checked again as them, even when the owner
461
+ clicks (the claim is their text; their link must still be live), so it never
462
+ runs with the owner's rules or System 1's auto-allows. The mode is decided once,
463
+ at the start, from their grant's mode there (`ParticipantMode.live_for`) and
464
+ saved on the child (`verify_fresh.mode`): `review` keeps every turn of it
465
+ inside `ReviewBoundary` (read only), `pair` inside `PairBoundary`.
466
+ `GuardrailWiring#turn_mode` reads the saved mode and never looks it up again,
467
+ and every turn of the child carries the participant as its author, whoever
468
+ started it, so the boundary can't fall away for a turn the owner typed into it
469
+ or for an author lookup that finds nobody (the child has no
470
+ `participants.json`; revoke and cap read the parent's). It fails closed: a
471
+ `verify_fresh` field that doesn't read is a participant nobody has, under
472
+ review, whose turns are refused. `ChildBoundary` applies on top, as for every
473
+ delegate child, and its asks go to the owner (once only, as a participant's
474
+ do; System 1 skips them). The owner's own fresh child is an ordinary delegate
475
+ child: the owner's turns, `ChildBoundary`.
476
+
455
477
  ## Rules in config.yml
456
478
 
457
479
  ```yaml
@@ -315,7 +315,8 @@ fields: a field added to an event later reaches nobody until it is named there.
315
315
  - **Events** (`.event`, the stream). `EVENTS` names the types a participant gets, each rebuilt by its own method:
316
316
  `turn_started` (turn id, its wall time `at`, the author and name, `mine` for their own, the prompt only when a person sent the turn:
317
317
  `ClientId.human?`, else `automatic`; `continue`), the turn's end (`turn_completed`, `turn_canceled`, `turn_failed`:
318
- type and turn id), `generation_chunk` (its visible `text` only; a thinking-only chunk is dropped),
318
+ type and turn id; `turn_completed` may also carry an allowlisted verdict card), `generation_chunk` (its visible
319
+ `text` only; a thinking-only chunk is dropped),
319
320
  `generation_retrying` (`restarted`), `tool_call_started` (the tool's name, and for a file tool the path relative to
320
321
  the session's folder when it is inside it; never when that folder is home or `/`), `tool_call_completed` (its status word), and bare markers (the type
321
322
  alone) for the queue (`turn_enqueued`, `turn_dequeued`, `input_merged`), the owner's question, approval or Continue
@@ -414,6 +415,72 @@ fields: a field added to an event later reaches nobody until it is named there.
414
415
  cookie is one name: a second participant link redeemed there replaces the first), and **the owner opening a
415
416
  participant's invite link** in their own browser swaps their owner cookie for that grant (the page asks first). A
416
417
  continue archives the session and revokes its grants: re-share, and links to the old session stay on it.
418
+ - **Verify verdict cards** (`verdict_card.js`). The built-in Verify preset carries the selected source turn id and quote
419
+ with the submitted turn, without changing its prompt text. A completed Verify reply whose final non-empty line has
420
+ a valid `Verdict: confirmed|refuted|unclear — <reason>` appends a verdict to the session. Both pages place its card
421
+ after the source turn, name its author, show the reason and quote, and link to the checking turn; a live completion
422
+ adds the card without a reload. The participant view receives only the card fields through its allowlist, with the
423
+ quote and reason redacted by the view's normal rules, never the full reply.
424
+ - **Verify fresh** (`VerifyFresh`, `verdict_fresh.js`). An `unclear` card has a **Verify fresh** button (the owner's
425
+ page always; a participant's while `can_comment`, i.e. while their grant could run a turn now; never in the owner's
426
+ participant view). It is `POST /api/sessions/:id/verdicts/:check_turn_id/fresh` (`VerifyFreshRoutes`, `:owner`
427
+ and `:participant`), which starts a **fresh child**: a delegate child (`spawn_session`, `delegate: true`,
428
+ `parent_id`) in the session's own folder (no worktree; a folder outside git too), on its model, whose first and only
429
+ prompt is the claim, quoted, and "Verify this claim against the files here. You have none of the original session's
430
+ context. End with a line `Verdict: confirmed|refuted|unclear — <one-line reason>`." No parent history. Its first
431
+ turn is a Verify turn on the same anchor, so its verdict is parsed. Refused: a verdict the session doesn't have (404
432
+ `not_found`; a fresh verdict itself is never the target, its original is), one that isn't unclear (409
433
+ `not_unclear`), and, checked together under `verify_fresh.lock`: a fresh check of that verdict still running (409
434
+ `fresh_running`), a fresh child of the same requester still running (409 `fresh_busy`: one at a time per
435
+ participant, and one for the owner; each start passes the cap on saved spend, so parallel children would multiply
436
+ it), and the parent already running `session.max_children` children of any kind (409 `children_full`, so fresh
437
+ checks never crowd out the owner's own delegates). For a participant also what refuses their turn: their cap (429
438
+ `cap_reached`, counting their fresh children) and a live worker that predates participants' turns (409
439
+ `review_unavailable`). A start that involves a participant is audited in the parent's trail (`verify_fresh`:
440
+ `participant` whose child it is, `asked_by` the clicking participant's id or `owner`, `of`, `status`, the child and
441
+ mode, or the reason); the answer is `{of, running: true}`, the owner's with `child_id` too.
442
+ - **Whose child it is.** A verdict a participant's turn recorded (`by: p:<id>`) is checked again **as that
443
+ participant**, whoever clicks, the owner included: the claim is their text (the quote their Verify turn sent),
444
+ so it never runs with the owner's power (the owner's rules, System 1's auto-allows). Their grant must still be
445
+ live in this session, else 409 `author_link_ended` ("Sam's link has ended: …"), and their cap and the worker's
446
+ features are checked as for their turn (a participant who clicks passes their own checks first). An owner's
447
+ verdict (`by` none) is checked again as whoever clicks: an owner's child for the owner, the clicking
448
+ participant's own for a participant.
449
+ - **The child is its requester's.** Its session file carries `verify_fresh: {author, mode, of, anchor}`
450
+ (`VerifyFreshOrigin`), saved before its worker starts. `author` is the participant's (`p:<id>`), none for the
451
+ owner. `mode` is the participant's grant's mode in the parent at that moment (`ParticipantMode.live_for`:
452
+ `pair` only for a grant stamped to act in a working session whose worker runs pair turns, else `review`). Every
453
+ turn of a participant's child is theirs, whoever starts it (the owner typing into it, a continue, a reminder):
454
+ `VerifyFreshOrigin#origin_for` sets the author in `Engine#begin_turn` and in the worker, so their boundary
455
+ (`GuardrailWiring#turn_mode` reads the saved mode, never a lookup), their System 1 skip, their once-only asks to
456
+ the owner and the stamps on their messages follow. The child has no `participants.json`: its participant lookups
457
+ (`Session#participants_source_id`) read the parent's, so a revoke, an expiry or the parent's archive refuses the
458
+ child's next turn and stops a pair turn mid-turn, as in the parent; their audit lines go to the parent's
459
+ `participants.jsonl`. A field that doesn't read is `VerifyFreshOrigin::INVALID`: a participant author nobody
460
+ has, under review, so its turns are refused, never run as the owner's. Its approvals wait on the child for the
461
+ owner (D2); the owner's page links to it.
462
+ - **What follows from that.** The owner typing into a participant's child is that participant's turn: charged to
463
+ their cap and run under their boundary (open an ordinary session to act with the owner's own). A revoke (or an
464
+ expiry, or the parent's archive) stops a pair-mode child mid-turn, as a pair turn; a review-mode child finishes
465
+ the turn it is in and is refused at its next turn's start, as a review turn. A continue of the parent archives it
466
+ (and revokes its grants) and moves a running fresh child to the new link with the other open delegates: the
467
+ turn it is in still forwards its verdict to the archived link (the parent it started under), which shows it
468
+ there, and its next turn is refused (its requester's grant isn't on the new link). Nothing reaches the new link's
469
+ card: fails closed.
470
+ - **Spend.** The parent's `verify_fresh.jsonl` (one line per child: `child_id`, `of`, `author`, `at`) is the
471
+ rollup's index: a participant's cap check in chi web (`#participant_saved_spend`), their usage line, and the
472
+ worker's (`Worker#participant_spend`) add their fresh children's saved spend (`VerifyFresh.children_spend`); a
473
+ child's worker adds the parent's saved spend and its siblings'. A child's spend counts from its turns' ends, as
474
+ a turn's does.
475
+ - **The verdict lands on the card**, never in the parent's conversation: the child's worker writes each verdict
476
+ its turn recorded into the parent's `fresh_verdicts/<check turn>.json` (`fresh: true`, `of`, `child_id`) after its
477
+ save, and a fresh child never rings its parent (no delegate report, `ChildRing.ring`). The parent's worker takes
478
+ them into its `verdicts` between turns (`Worker#absorb_fresh_verdicts`), saves, deletes the files and announces
479
+ `:verdict_recorded`; chi web shows the ones still waiting (`VerifyFresh.verdicts_of`: the summary, the review
480
+ data). The card shows "Verifying fresh…" while the child runs (the summary's `verify_fresh` `{of, child_id}`,
481
+ the review data's `fresh_running`, ids only) and the fresh verdict as a `fresh: ✅ confirmed — …` line on the same
482
+ card. The participant view passes `fresh` and `of`, never the child's id (a participant can't open it). No push
483
+ for a fresh verdict yet (`Attention` pushes a verdict at a turn's end).
417
484
  - **The comment box** is the owner's composer (`composer_mount.js`, styled by the shared `composer.css`), mounted by
418
485
  `participant.js` with a subset on. It has: the height that grows with the text, the grip (drag to resize,
419
486
  double-click resets, click or drag down folds it; folded, the grip is the handle on the dock's edge and typing on the
@@ -532,7 +599,8 @@ invited to has none: its files stay byte for byte as they were.
532
599
  remembered and audited; or `what: command`, a participant's `/word` or
533
600
  `!line` the Bridge or the worker refused), `turn` (chars, `status: accepted` with its enqueued id, or `refused` with
534
601
  the reason: `not_permitted`, `too_many_turns`, `cap_reached`, `review_unavailable`, `worker_timeout`), `cap` (a new
535
- cap). It goes with the session (retention, delete).
602
+ cap), `verify_fresh` (`of`, `status`, the child and mode or the reason). A Verify fresh child's participants' lines
603
+ are its parent's. It goes with the session (retention, delete).
536
604
  - **`participants.lock`** (0600, empty): the lock a participant's turn file is counted and written under when no
537
605
  worker runs ([Turns](#turns)).
538
606
 
@@ -1,7 +1,8 @@
1
1
  # Web Push (internals)
2
2
 
3
3
  chi web pushes to the owner's phone when a session needs them, for what the tab notifies for: an approval, a
4
- question, a step-limit question, a card with actions, a failed turn, and (per device, off by default) a done one. The
4
+ question, a step-limit question, a card with actions, a failed turn, a recorded Verify verdict, and (per device,
5
+ off by default) a done one. The
5
6
  phone runs chi web as a Home Screen app (iOS 16.4+), its bell subscribes the device, and chi web, the only sender,
6
7
  pushes through the device's push service (Apple's `web.push.apple.com`). The user side is in
7
8
  [configuration.md](../configuration.md#phone-notifications-web-push). This page is how it is built (P1, P2a and P2b of the
@@ -146,17 +147,20 @@ the top and in `pending_question` / `pending_card`, strings in `last_turn`), `no
146
147
  3.5 s (notify.js `RELAY_HOLD_MS`): its parent may relay it.
147
148
  2. a new `pending_card` id: `card`, key `"<sid>:card:<cid>"`.
148
149
  3. a top-level turn whose `last_turn["ended_at"]` changed (a delegated child's is its parent's to report):
149
- `failed`, or `done` when `completed` and `seconds` ≥ `DONE_MIN_SECONDS` 10 or its origin is in
150
- `UNWATCHED_ORIGINS` (`delegate_report`, `context`, `task`); key `"<sid>:<ended_at>"`. A reminder turn follows
151
- the same rules (no reminder event of its own).
150
+ a newly recorded verdict is `verdict`, keyed by `"<sid>:<check_turn_id>"`; otherwise `failed`, or `done` when
151
+ `completed` and `seconds` ≥ `DONE_MIN_SECONDS` 10 or its origin is in `UNWATCHED_ORIGINS`
152
+ (`delegate_report`, `context`, `task`), keyed by `"<sid>:<ended_at>"`. A verdict wins over the done/no-event
153
+ rule, including for a short turn. A reminder turn follows the same rules (no reminder event of its own).
152
154
  - `WAITING` (approval, question, continue, card) are the reasons that stay open until answered: badged, in `open`,
153
- cleaned up, and pushed only while still open. A failed or done turn has nothing to answer.
155
+ cleaned up, and pushed only while still open. A failed, done or verdict turn has nothing to answer.
154
156
  - `Attention.closed_keys(prev, next)`: the question answered, replaced, newly relayed; the card resolved or
155
157
  replaced; both when the session is gone.
156
158
  - `Attention.text(summary, reason)`: notify.js `attentionText`'s words: the session's name as its card has it
157
159
  (title, auto title, the recap's first sentence, first preview, last prompt — the same order `format.js cardName`
158
160
  uses) clipped to 61, and "needs approval" / "needs an answer" / "hit its step limit" / "needs you" / "turn
159
- failed" / "done in N s" (`last_turn["seconds"]` rounded). Never the question or the command.
161
+ failed" / "done in N s" (`last_turn["seconds"]` rounded) / "<who>: <icon> <verdict> — <reason>" (reason clipped
162
+ to 120 characters). A verdict names its participant from the session's participant sidecar, or says "You" for the
163
+ owner; only verdicts use that push copy, never the full reply. Never the question or the command.
160
164
 
161
165
  `Notifier#call(event)` is a `SessionHub` sink. **The hub calls it under its Monitor**, from the tick thread and also
162
166
  from chi web's request threads (`SessionHub#touch`, from the stop/archive/delete routes), so it only enqueues: a push
data/docs/sessions.md CHANGED
@@ -4,6 +4,8 @@ Sessions are plain files — no DB. Each session is `~/.local/state/samagotchi/s
4
4
 
5
5
  **Message times:** an entry of `messages` may carry `at`, when it happened: a local-offset ISO time with milliseconds (`2026-10-24T14:32:05.120+02:00`). User prompts, lines merged into a turn and steers, reminders, context and task notes, and the model's replies have one (a note's is its `created_at`); tool results, `!cmd` messages and chi's own turn notes don't, nor does anything saved before chi stamped messages (no backfill). It is for the UIs only (the gap dividers, a bubble's time): it never goes to the model, so it changes nothing in the requests or their prompt cache. The one time the model reads is the reminders': due reminders go into a turn as a system message headed `[SYSTEM: REMINDERS DUE 2026-10-24 14:32 local]` (the turn's local time; `ReminderQueue#inject!`), written into the saved content then, so a later request sends the same text.
6
6
 
7
+ **Verify verdicts:** a web Verify turn stamps its opening user message with `verify: {turn_id, quote}` (the selected source turn and quote; metadata only, never added to the prompt). When that turn completes and the final assistant reply's last non-empty line is a valid `Verdict: confirmed|refuted|unclear — <reason>`, the session file appends `{anchor: {turn_id, quote}, verdict, reason, by, check_turn_id, at}` to its optional `verdicts` array. `by` is the turn's author (nil for the owner), `check_turn_id` names the Verify turn, and `at` is its timestamp. The array is append-only; older files without it have no saved verdicts. An `unclear` verdict can be checked again by **Verify fresh** (a delegate child that gets only the claim; see *Delegating*): its verdict is appended to the parent's `verdicts` with `fresh: true`, `of` (the verdict it checked again) and `child_id`, from the parent's `fresh_verdicts/` dir, which its worker takes between turns. Verdicts are display state: none enters a prompt.
8
+
7
9
  **Input files:** a message for a session's worker waits in `input/<timestamp>.json` (`SessionInbox.write_input`: `prompt`, `client_id`, `enqueued_id`, and `images`, `no_interrupt` and `delivery` when they apply; `delivery` is written only for `"queue"`). While a turn runs, the worker merges them at each step boundary in name order, except a queued one or one with images: those wait for the turn's end and run one turn each, in name order, and a file after them still merges (`Worker#pending_input_drain`). A queue is sent only to a worker whose `bridge.json` names the `delivery` feature; an older one would steer it, so `SessionManager.deliver_turn` refuses it (409 `delivery_unavailable`), and any other 4xx the Bridge answers is a refusal too, never a fallback file (except a 413, a body over its size limit: that one still goes in as a file, its `delivery` kept). A session command recognized in a file runs without a turn; its outcome is kept at `input/settled/<enqueued_id>` (`completed`, `failed` for a command error, `canceled` with `cancel_reason` `busy` or `dropped` for one refused mid-turn or dropped as the worker left; `origin: command`) so `chi send --wait` ends instead of waiting for a turn that will not run, its detail saying `the command failed` or `the command was refused (busy)` rather than a turn's words. **Reply files:** a turn that ends with visible text leaves its reply in `output/<timestamp>.txt`, or `output/<timestamp>-<enqueued_id>.txt` when the turn ran an input file with an id (`SessionInbox.write_output`), and the session's `last_turn` keeps that `enqueued_id`; so a wait for one queued message (`chi send --queue --wait`, `ReplyWait`'s `turn:`) finds its own turn's reply or end, never another turn's. The names still sort by time, and every reader takes `*.txt`.
8
10
 
9
11
  **Removing a queued message:** a message queued for after the running turn (`chi send --queue`, the web's Alt+Enter, `/queue TEXT` in a terminal), or one with images (it waits for the turn's end too), can be taken back while it still waits: the worker's Bridge (`POST /session/:id/turn/dequeue {enqueued_id}`, the `unqueue` feature) deletes its input file and every UI drops it (`:turn_dequeued`). It is a race with the turn's end, decided by the same rename the worker uses to take the file: once a turn took it the answer is `409 already_started` (it runs), and an id the worker no longer knows is `404 unknown_enqueued_id`. A plain message sent without `--queue` is not removable (`409 not_queued`): a running turn's next step may take it at any moment. A wait for its turn (`chi send --queue --wait`, a delegate follow-up's wait) ends at once with `removed` (the tombstone `input/removed/<enqueued_id>` says so), and a `chi -p` run whose queued prompt went ends with exit 1. The `200` carries the removed `prompt`: the web and the attached terminal put it back into the composer to edit (not its images). The images a removed message named stay in the session's `images/`. Not removable: a queued session command (`/queue /model x`: it waits in the worker, not in `input/`), a message in a stopped worker's inbox (the next worker runs it), and the plain REPL's (`--no-shared`) own queue.
@@ -384,6 +386,7 @@ The `delegate` tool hands a task to a **child session**: an ordinary chi session
384
386
  - Every child starts with the shipped `delegated` system memory preloaded (`preloaded_memory_names: ["system/delegated"]`; `--mute delegated` would hide it): put everything in the final reply, don't delegate further, don't write memories or change config, stay in the folder. Its system prompt also says `Delegated by session <parent id>`. The parent's own `--mute` list does not carry over.
385
387
  - **A child asks before it changes anything outside its folder** (the work tree top of the folder it started in): a write or edit there, mutating git in another checkout, a command that isn't read-only run from outside it or naming the main checkout or a sibling worktree. It's a core guardrail, on in every mode and without the guardrails bundle, and answered once or for the child's session ([Guardrails](guardrails.md#a-delegate-childs-boundary)); the approval relays to the parent's user as any other.
386
388
  - Limits: a child cannot `delegate` (depth 1; the tool refuses), and one session may have at most `session.max_children` (default `4`, env `SAMAGOTCHI_SESSION_MAX_CHILDREN`) children running at a time, counted from the session files (a finished or idle-exited child does not count; one whose worker is still starting does, for 15 s). A refused call creates no session. Nothing stops children with the parent: `chi sessions stop ID` does.
389
+ - **Verify fresh children** (a web card's Verify fresh, [Participants: Verify fresh](internals/participants.md)) are delegate children chi web starts, not the model: in the parent's own folder, with only the claim as their task and none of the parent's history. Their session file carries `verify_fresh: {author, mode, of, anchor}`; a participant's child runs every turn as that participant's (their boundary, saved at the start; their cap, counted with the parent's spend; the parent's `participants.json`). One never rings its parent: its verdict goes to the parent's verdict card (`fresh_verdicts/`), not its conversation, and the parent's `verify_fresh.jsonl` lists them for the spend rollup and the running check (one per verdict at a time). They count among the parent's `session.max_children` while they run.
387
390
  - The link is the session's `parent_id`, written before the worker starts, so a respawn keeps it (and `delegate: true`, which a fork's `parent_id` doesn't come with; a child saved before that field is known by its `system/delegated` memory); `GET /api/sessions`, `/api/sessions/:id` and the web hub's `session` frames carry it and `delegate` (a fork is `false`; the web parent's children chip counts delegates only); `SessionManager.children_of` lists a parent's children.
388
391
  ## Restart fallback
389
392
 
@@ -1071,6 +1071,7 @@ module Samagotchi
1071
1071
  # request's principal; the Bridge is loopback and trusts its caller).
1072
1072
  author = fetched(parsed, "author")
1073
1073
  author = nil unless author.is_a?(String) && !author.empty?
1074
+ verify = VerifyAnchor.from(fetched(parsed, "verify"))
1074
1075
  if sid.to_s.strip.empty? || prompt.to_s.strip.empty?
1075
1076
  return [{ "Allow" => "POST" }, 400,
1076
1077
  { error: "missing_fields", detail: "session_id and prompt are required" }]
@@ -1132,7 +1133,7 @@ module Samagotchi
1132
1133
  delivery = Delivery::QUEUE if @accumulator.participant_turn?
1133
1134
  queued_behind_turn = Delivery.queue?(delivery) && !@accumulator.current_turn.nil?
1134
1135
  enqueue_turn(prompt: prompt, client_id: client_id, enqueued_id: enqueued_id,
1135
- no_interrupt: no_interrupt, images: images, delivery: delivery, author: author).tap do |ok|
1136
+ no_interrupt: no_interrupt, images: images, delivery: delivery, author: author, verify: verify).tap do |ok|
1136
1137
  next unless ok
1137
1138
 
1138
1139
  enqueued_event = { type: :turn_enqueued, enqueued_id: enqueued_id, client_id: client_id, prompt: prompt.to_s }
@@ -1296,9 +1297,9 @@ module Samagotchi
1296
1297
 
1297
1298
  # Write a turn into this session's input dir, reusing the file IPC the
1298
1299
  # worker polls. Never calls run_turn across the boundary.
1299
- def enqueue_turn(prompt:, client_id: nil, enqueued_id: nil, no_interrupt: false, images: [], delivery: nil, author: nil)
1300
+ def enqueue_turn(prompt:, client_id: nil, enqueued_id: nil, no_interrupt: false, images: [], delivery: nil, author: nil, verify: nil)
1300
1301
  SessionInbox.write_input(session_dir, prompt: prompt, client_id: client_id, enqueued_id: enqueued_id,
1301
- no_interrupt: no_interrupt, images: images, delivery: delivery, author: author)
1302
+ no_interrupt: no_interrupt, images: images, delivery: delivery, author: author, verify: verify)
1302
1303
  end
1303
1304
 
1304
1305
  # ── HTTP plumbing ────────────────────────────────────────────────────────
@@ -14,6 +14,10 @@ require_relative "bridge_client/event_stream"
14
14
  require_relative "bridge_client/stream_request"
15
15
 
16
16
  module Samagotchi
17
+ # Loaded on first use (a Verify turn): keeps the bundle test kit's load
18
+ # budget (spec/samagotchi/testing/sandbox_spec.rb).
19
+ autoload :VerifyAnchor, File.expand_path("verify_anchor", __dir__)
20
+
17
21
  # Client side of a session worker's Bridge: the 127.0.0.1 HTTP + SSE server
18
22
  # each SessionManager worker runs for its Engine (see Bridge). Discovery goes
19
23
  # through WorkerLink (the worker's `bridge.json` sidecar, WorkerSidecar, in
@@ -182,13 +186,16 @@ module Samagotchi
182
186
  # owner; sent only when given (SessionManager.deliver_turn)
183
187
  # @param pending_limit [Integer, nil] with a participant's +author+: how
184
188
  # many of their turns may be running or waiting (429 too_many_turns past it)
185
- def post_turn(prompt:, client_id: nil, no_interrupt: false, images: nil, delivery: nil, author: nil, pending_limit: nil)
189
+ def post_turn(prompt:, client_id: nil, no_interrupt: false, images: nil, delivery: nil, author: nil, pending_limit: nil,
190
+ verify: nil)
186
191
  body = { session_id: @session_id, prompt: prompt, client_id: client_id }
187
192
  body[:no_interrupt] = true if no_interrupt
188
193
  body[:author] = author if author
189
194
  body[:pending_limit] = pending_limit if author && pending_limit
190
195
  body[:images] = images if images && !images.empty?
191
196
  body[:delivery] = delivery unless delivery.nil? || Delivery.next_step?(delivery)
197
+ anchor = VerifyAnchor.from(verify)
198
+ body[:verify] = anchor.to_file if anchor
192
199
  body[:deadline] = deadline
193
200
  post("turn", body, read_body: true)
194
201
  end
@@ -26,6 +26,8 @@ class KnownNames
26
26
  # name after it is what is checked.
27
27
  TOKEN_SPLIT = %r{[/\s"'=:,;|&()<>@~#]+}
28
28
  PART_SPLIT = /[-_.]+/
29
+ # Separators at a token's start or end, dropped before it is compared.
30
+ EDGE_SEPARATORS = /\A[-_.]+|[-_.]+\z/
29
31
  # A candidate with a shell glob character is not checked: `workato*`
30
32
  # is a pattern over the name, not a misspelling of it.
31
33
  GLOB = /[*?\[\]{}]/
@@ -94,15 +96,22 @@ class KnownNames
94
96
  return nil if names.empty?
95
97
 
96
98
  scan_targets(event).each do |where, text|
97
- candidates(text).each do |candidate|
99
+ candidates(text).each do |token|
100
+ next if names.key?(token.downcase)
101
+
102
+ # "coffee_assistant-" (a prefix search) is the name, not a misspelling.
103
+ candidate = token.gsub(EDGE_SEPARATORS, "")
98
104
  down = candidate.downcase
99
- next if names.key?(down)
105
+ next if candidate.empty? || names.key?(down)
100
106
 
101
107
  names.each do |name_down, name|
108
+ next if camel_case?(candidate, name)
109
+
102
110
  max = @max_distance || (name.length < 10 ? 1 : 2)
103
- next if (candidate.length - name.length).abs > max
111
+ bare = name_down.gsub(EDGE_SEPARATORS, "")
112
+ next if (candidate.length - bare.length).abs > max
104
113
 
105
- d = self.class.distance(down, name_down)
114
+ d = self.class.distance(down, bare)
106
115
  return { miss: candidate, name: name, distance: d, where: where } if d.between?(1, max)
107
116
  end
108
117
  end
@@ -110,6 +119,18 @@ class KnownNames
110
119
  nil
111
120
  end
112
121
 
122
+ # "CoffeeAssistant" or "coffeeAssistant" for "coffee_assistant": the
123
+ # name's own class or variable form, not a misspelling of it. Only a
124
+ # form with a capital: "coffeeassistant" and "coffee-assistant" are
125
+ # still near misses.
126
+ def camel_case?(candidate, name)
127
+ parts = name.split(PART_SPLIT)
128
+ return false if parts.size < 2
129
+
130
+ upper = parts.map(&:capitalize).join
131
+ [upper, upper[0].downcase + upper[1..]].include?(candidate)
132
+ end
133
+
113
134
  # [where, text] pairs to scan: the command, the cwd a command names,
114
135
  # each path, and a write/edit call's own path (in case normalisation
115
136
  # dropped it). Not file contents.
@@ -1,6 +1,6 @@
1
1
  ---
2
2
  name: known-names
3
- version: 0.1.6
3
+ version: 0.1.7
4
4
  scope: system
5
5
  description: Catches a near-miss of a protected name (home folder, login, git name, repo name, names from config) in a tool call and rejects it with the right spelling; mode ask / correct optional
6
6
  trust_level: reviewed
@@ -8,7 +8,7 @@ files:
8
8
  known_names.md: sha256:d5348b95b6c655017b541910b36e384220da280818160201f6ec9a50e1e4287c
9
9
  hooks:
10
10
  known_names.rb:
11
- sha256: sha256:7743adea25655d019d308793e7616043fc29e3d664e1572b50d0b27bf3a9238f
11
+ sha256: sha256:a851637e9292f825dc5b8877cd060184de976aa4297b114dab412a89fa3cf091
12
12
  event: before_tool_call
13
13
  on_error: log
14
14
  priority: 50
@@ -1,6 +1,6 @@
1
1
  ---
2
2
  name: samagotchi-system
3
- version: 0.125.0
3
+ version: 0.126.0
4
4
  scope: system
5
5
  description: Default system memories — identity, self map, config modification protocol, memory guide and the delegated-session rules
6
6
  files:
@@ -64,6 +64,9 @@ module Samagotchi
64
64
  def ring(child, why:, state_dir:, wake: nil)
65
65
  return nil if mode == "off"
66
66
  return nil unless child&.delegate?
67
+ # A Verify fresh child's verdict reaches its parent's card
68
+ # (VerifyFresh.forward), never its conversation.
69
+ return nil if child.verify_fresh
67
70
 
68
71
  parent_id = Session.parent_override(child.id, state_dir: state_dir) || child.parent_id
69
72
  return nil unless parent_id
@@ -86,11 +86,16 @@ module Samagotchi
86
86
  state: state_of(session, row, reply), waiting: row[:waiting], live: !!row[:live],
87
87
  delegate: session.delegate?, cwd: session.working_directory, branch: GitHead.branch(session.working_directory),
88
88
  last_reply: reply && first_line(session.id, reply, state_dir: state_dir),
89
- last_reply_at: reply && reply_time(reply), reported: !reply.nil? && cursor.reply_file == reply,
89
+ last_reply_at: reply && reply_time(reply), reported: !reply.nil? && reported?(session, cursor, reply),
90
90
  updated_at: session.updated_at, archived: !!row[:archived])
91
91
  end
92
92
  private_class_method :child
93
93
 
94
+ # A reply the parent was given; a Verify fresh child's always counts
95
+ # (its verdict goes to the parent's card, never a report).
96
+ def reported?(session, cursor, reply) = !session.verify_fresh.nil? || cursor.reply_file == reply
97
+ private_class_method :reported?
98
+
94
99
  def state_of(session, row, reply)
95
100
  return "waiting" if row[:waiting]
96
101
  return "running" if Tools::Delegate.running?(row)
@@ -15,6 +15,8 @@ require_relative "steer"
15
15
  require_relative "turn_note"
16
16
  require_relative "turn_endings"
17
17
  require_relative "turn_origin"
18
+ require_relative "verify_anchor"
19
+ require_relative "verdict"
18
20
  require_relative "model_profile"
19
21
  require_relative "effective_model"
20
22
  require_relative "session_setting"
@@ -901,10 +903,12 @@ module Samagotchi
901
903
 
902
904
  # Event types #announce accepts: facts about the session's input queue,
903
905
  # a failed turn's prompt handed back and the continue offer, which live
904
- # UIs need in the event log, emitted outside any turn's stream.
906
+ # UIs need in the event log, emitted outside any turn's stream; a
907
+ # Verify fresh child's verdict landing on its card (:verdict_recorded).
905
908
  ANNOUNCEABLE_EVENTS = %i[turn_enqueued turn_dequeued input_merged prompt_restored continue_offered continue_resolved
906
909
  command_queued command_ran context_added card hook_notice guardrail_warning
907
- plugin_init_started plugin_init_finished working_set_updated participants_changed].freeze
910
+ plugin_init_started plugin_init_finished working_set_updated participants_changed
911
+ verdict_recorded].freeze
908
912
 
909
913
  # Put a transport-level event into the ordered event log. Unlike turn
910
914
  # events it reaches only persistent observers (no turn sink, no memory
@@ -1589,7 +1593,10 @@ module Samagotchi
1589
1593
  seconds = turn.elapsed
1590
1594
  synchronize_events do
1591
1595
  kept, event = yield(seconds)
1596
+ stamp_verify_anchor(kept, turn)
1592
1597
  replace_session_messages(turn.session, kept) if kept
1598
+ verdict = record_verdict(turn, outcome)
1599
+ event[:verdict] = verdict if verdict
1593
1600
  emit_working_set(turn.on_event, turn.session, turn.session.messages)
1594
1601
  turn.session.status = Session::STATUS_IDLE
1595
1602
  record_last_turn(turn.session, outcome, seconds, turn.origin, event, limit: exhausted ? turn.limit : nil)
@@ -1602,6 +1609,31 @@ module Samagotchi
1602
1609
  end
1603
1610
  private :end_turn
1604
1611
 
1612
+ def stamp_verify_anchor(messages, turn)
1613
+ return unless turn.verify && messages.is_a?(Array)
1614
+
1615
+ opening = messages.reverse.find do |message|
1616
+ (message[:role] || message["role"]).to_s == "user" &&
1617
+ (message[:turn_id] || message["turn_id"]).to_s == turn.id
1618
+ end
1619
+ opening[:verify] = turn.verify.to_h if opening
1620
+ end
1621
+ private :stamp_verify_anchor
1622
+
1623
+ def record_verdict(turn, outcome)
1624
+ return unless outcome == "completed" && turn.verify
1625
+
1626
+ answer = AnswerTail.find(turn.session.messages, turn_id: turn.id)
1627
+ parsed = Verdict.parse(answer && (answer[:content] || answer["content"]))
1628
+ return unless parsed
1629
+
1630
+ record = { anchor: turn.verify.to_h, verdict: parsed.verdict, reason: parsed.reason,
1631
+ by: turn.origin.author, check_turn_id: turn.id, at: Time.now.iso8601(3) }
1632
+ turn.session.verdicts = turn.session.verdicts + [record]
1633
+ record
1634
+ end
1635
+ private :record_verdict
1636
+
1605
1637
  def save_quietly(session)
1606
1638
  session.save(state_dir: session_state_dir)
1607
1639
  rescue StandardError => e
@@ -2011,7 +2043,7 @@ module Samagotchi
2011
2043
  # +wall+: the wall time it started (MessageTime.now, once): the
2012
2044
  # :turn_started event's at and its user message's (and reminders') stamp.
2013
2045
  Turn = Struct.new(:session, :prompt, :continue, :on_event, :controller, :origin, :started_at, :messages, :ended,
2014
- :settings, :id, :limit, :resumed, :resume_from, :wall) do
2046
+ :settings, :id, :limit, :resumed, :resume_from, :wall, :verify) do
2015
2047
  def resumed? = resumed ? true : false
2016
2048
 
2017
2049
  # The turn's boundary events carry the origin only when it names a
@@ -2083,9 +2115,9 @@ module Samagotchi
2083
2115
  # re-raises; a provider error adds error_kind:, retryable:, host: and a
2084
2116
  # one-line summary:.
2085
2117
  def run_turn(session, prompt, on_event: nil, max_iterations: nil, cancel_controller: nil, max_tool_output_chars: nil, pending_input: nil, continue: false,
2086
- origin: TurnOrigin::NONE, images: [], id: nil, resumed: false, resume_from: nil)
2118
+ origin: TurnOrigin::NONE, images: [], id: nil, resumed: false, resume_from: nil, verify: nil)
2087
2119
  turn = begin_turn(session, prompt, on_event: on_event, cancel_controller: cancel_controller, origin: origin,
2088
- continue: continue, id: id)
2120
+ continue: continue, id: id, verify: verify)
2089
2121
  if continue
2090
2122
  turn.resumed = resumed
2091
2123
  turn.resume_from = resume_from
@@ -2133,8 +2165,11 @@ module Samagotchi
2133
2165
  # The turn's state before anything can fail it: the session running,
2134
2166
  # the turn flag, its cancel controller and sink. The cross-thread state
2135
2167
  # is set here and cleared in #release_turn only.
2136
- def begin_turn(session, prompt, on_event:, cancel_controller:, origin:, continue:, id: nil)
2168
+ def begin_turn(session, prompt, on_event:, cancel_controller:, origin:, continue:, id: nil, verify: nil)
2137
2169
  origin = TurnOrigin.of(origin)
2170
+ # A participant's Verify fresh child: every turn is theirs (their
2171
+ # author, boundary and cap), whoever started it.
2172
+ origin = session.verify_fresh.origin_for(origin) if session.respond_to?(:verify_fresh) && session.verify_fresh
2138
2173
  # Track the active session for recap and status snapshot.
2139
2174
  @session = session
2140
2175
  # status is turn state: running now, idle again before the turn's end
@@ -2157,7 +2192,7 @@ module Samagotchi
2157
2192
 
2158
2193
  Turn.new(session: session, prompt: continue ? nil : prompt, continue: continue, on_event: on_event,
2159
2194
  controller: effective_controller, origin: origin, started_at: Process.clock_gettime(Process::CLOCK_MONOTONIC),
2160
- id: id || SecureRandom.uuid, wall: MessageTime.now)
2195
+ id: id || SecureRandom.uuid, wall: MessageTime.now, verify: VerifyAnchor.from(verify))
2161
2196
  end
2162
2197
  private :begin_turn
2163
2198
 
@@ -2360,6 +2395,7 @@ module Samagotchi
2360
2395
  unless turn.continue
2361
2396
  user_message = { role: "user", content: turn.prompt, turn_id: turn.id, **author_fields(turn.origin.author),
2362
2397
  **participant_mode_fields(turn.origin.author), MessageTime::KEY => turn.at }
2398
+ user_message[:verify] = turn.verify.to_h if turn.verify
2363
2399
  user_message[:images] = image_refs unless image_refs.empty?
2364
2400
  turn.messages << user_message
2365
2401
  session.last_prompt = turn.prompt
@@ -109,9 +109,18 @@ module Samagotchi
109
109
  @git = Guardrails::GitInfo.new
110
110
  author = turn_author
111
111
  @turn_entry = author && @participant.call(author)
112
- @turn_mode = author && ParticipantMode.for_turn(@session_lookup.call, @turn_entry)
112
+ @turn_mode = author && participant_turn_mode(@session_lookup.call, @turn_entry)
113
113
  end
114
114
 
115
+ # A participant's turn's mode: in a Verify fresh child the one saved
116
+ # on it at spawn (VerifyFreshOrigin#turn_mode: :review unless saved
117
+ # :pair), never looked up again; elsewhere ParticipantMode.for_turn.
118
+ def participant_turn_mode(session, entry)
119
+ fresh = session.respond_to?(:verify_fresh) && session.verify_fresh
120
+ fresh ? fresh.turn_mode : ParticipantMode.for_turn(session, entry)
121
+ end
122
+ private :participant_turn_mode
123
+
115
124
  # The running participant turn's mode (ParticipantMode.for_turn: :pair
116
125
  # only in a working session for a participant invited to act), decided
117
126
  # at its start; nil for the owner's turns. The boundary in #checks, the
@@ -447,6 +447,7 @@ module Samagotchi
447
447
  message[:images] = entry[:images] if entry[:images].is_a?(Array) && !entry[:images].empty?
448
448
  message[:turn_id] = entry[:turn_id] if entry[:turn_id]
449
449
  message[:turn_start] = true if entry[:turn_start] == true
450
+ message[:verify] = entry[:verify] if entry[:verify].is_a?(Hash)
450
451
  message[:thinking] = entry[:thinking] if entry[:thinking].is_a?(String) && !entry[:thinking].empty?
451
452
  message[:anthropic_content] = entry[:anthropic_content] if entry[:anthropic_content].is_a?(Array)
452
453
  ToolResponse::SAVED_KEYS.each { |key| message[key] = entry[key] if entry[key] }