@awebai/oats 0.46.0 → 0.47.0

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.
@@ -121,24 +121,30 @@ The qualified channel frame starts on its accepted option and permits one
121
121
  Enter only. Folder-trust and API-key questions remain unexpected and block.
122
122
  A qualified normal-output boundary closes observation without asserting readiness.
123
123
 
124
- **Qualified completion and limitation:** exact captured banners cover Claude
125
- 2.1.289 on darwin-arm64 at 110x35 with aweb-channel 1.7.11 installed, and the
126
- original plugin-absent banner. Both name Opus 5.5, API Usage Billing, auto mode
127
- and medium effort. Source-derived synthetic variants additionally allow exactly
128
- `Claude Max`, `Claude Pro`, `Claude Team` or `Claude Enterprise` in place of
129
- `API Usage Billing`; these are not observed subscription captures. Claude may
130
- prefer an account-specific plan display name over those fallback labels; other
131
- display names remain unqualified even for subscribers on those plans. Every other
132
- row stays exact apart from the bounded canonical-home directory line. This
133
- interim extension does not cover task-filled screens or bypass-permissions mode.
134
- Other model, billing, effort or permission banners remain unqualified: the one
135
- Enter can be submitted before `E_SPAWN_INCOMPLETE` with
136
- `launchPrompts.status: blocked`. The submitted answer remains in `answers` and
137
- the event receipt; the pane may already be active. Inspect the retained target;
138
- do not automatically retry input or relaunch it. The installed capture used an
139
- isolated configuration and dummy credentials with runtime network denied. It
140
- proves bounded UI completion, not live-account admission, connection or native
141
- receive readiness, and does not establish success for every production setup.
124
+ **Qualified completion and limitation:** the existing exact captured and
125
+ source-derived empty API/subscription frames remain supported. After one recorded
126
+ submitted channel Enter, the same qualified Claude 2.1.289 darwin-arm64 executable
127
+ also recognizes the [bounded bottom input/footer structure](launch-prompt-completion.md)
128
+ at 110x35. Its six mode labels, five effort labels and finite hints are
129
+ source-qualified; header/billing/model and preceding task text are not structural
130
+ completion conditions. The 21 exact question markers take precedence over both
131
+ completion paths after submission. Normal `? for shortcuts` and `esc to interrupt`
132
+ are not markers. Input authorization and exact-home consent are unchanged.
133
+
134
+ **Accepted reporting risk:** an unknown dialog retaining the composer/footer can
135
+ be misclassified completed. Completion permanently closes input authority, but
136
+ proves neither absence of all questions nor idle/account/channel readiness.
137
+ Missing effort, trailing notices, wrapped footers and other unsupported shapes
138
+ still block. A feature-flag footer change can reduce matches without a binary
139
+ change. A matched structural frame must be saved privately before completed is
140
+ audited; failure reports incomplete with the answer and target preserved.
141
+
142
+ A submitted Enter can therefore precede `E_SPAWN_INCOMPLETE`; the pane may already
143
+ be active. Inspect the answer, checked event receipt and retained target, rather
144
+ than automatically retrying input or relaunching. Existing API fixtures used
145
+ isolated dummy credentials; the observed Max regression is a labeled home-only
146
+ sanitized derivative of one controller-compared in-progress frame, not a new
147
+ end-to-end launch proof or general production-success claim.
142
148
 
143
149
  When a harness update changes the version-bound prompt frame, opted-in homes
144
150
  block again with `blocked: unexpected prompt` until fixtures are refreshed.
@@ -1626,13 +1626,15 @@ allocated. Use runtime inspection and the existing start/restart lifecycle.
1626
1626
  Same-key spawn replay cannot convert a blocked outcome to successful spawn.
1627
1627
 
1628
1628
  `answers` can contain a `submitted` answer even when the final status is
1629
- `blocked`: the accepted installed-plugin and plugin-absent completion banners
1630
- pin specific model/effort/permission text and finite billing labels, including
1631
- source-derived Max/Pro/Team/Enterprise variants of the empty API frame. A different normal-looking banner can block after Enter
1629
+ `blocked`: even after exact or structural completion recognition is qualified,
1630
+ unsupported input/footer shapes and known question markers can block after Enter
1632
1631
  while the actual pane is active. Consumers must preserve that submitted receipt
1633
1632
  and never translate this outcome into “nothing started” or automatic retry.
1634
1633
  The [completion limitation](configuration.md#exact-home-launch-prompt-consent)
1635
- is part of the shipping contract.
1634
+ is part of the shipping contract. Structural completion closes observation only;
1635
+ unknown dialogs retaining the recognized composer/footer can be misclassified.
1636
+ Mandatory structural evidence-save failure reports checked `incomplete` with the
1637
+ same answer and target preserved. Private receipt v2 adds no public DTO fields.
1636
1638
 
1637
1639
  Strict Desktop/pi readers must accept these additive fields and the new event
1638
1640
  kind before a writer containing them ships. For operator recovery, see
package/docs/desktop.md CHANGED
@@ -294,12 +294,12 @@ for a password or key, and never runs anything over ssh itself.
294
294
  ## Instances on servers
295
295
 
296
296
  Every instance a registered server reports shows in its workspace's roster, whoever spawned it. You
297
- can open its terminal, start, restart, stop and remove it, and read its readiness, activity, Git
297
+ can open its terminal, start, restart, stop and retire it, and read its readiness, activity, Git
298
298
  and diffs, as for a local one. The context panel's Soul tab and its Messaging & Teams list read
299
299
  through the server too (`oats inspect --server <id> --home <path>`), and its Work card, "model
300
300
  from" line and "older build" chip show the facts the server relays. Only its pull request is not read here, since the forge reads this
301
301
  computer's clones. Every command goes to the server by the instance's home (`--server <id> --home
302
- <path>`), never by a bare name. Stop and Remove show the plan the server makes, and confirm
302
+ <path>`), never by a bare name. Stop and Retire show the plan the server makes, and confirm
303
303
  against it.
304
304
 
305
305
  A read waits for the server: the view says "Reading from <server>…", and gives up after about
@@ -615,7 +615,8 @@ anchor edited or removed since the original pre-hook snapshot, instead of
615
615
  replacing concurrent changes with stale metadata. `launchPrompts` records `status`,
616
616
  `answers`, `reason` and event `receipt`; unknown prompts use
617
617
  `blocked: unexpected prompt`. Status `incomplete` means the audit could not be
618
- completed, including a possible key followed by an audit failure. It is
618
+ completed, including a possible key followed by an audit failure or failure to
619
+ retain mandatory structural-completion evidence. It is
619
620
  not proof that no input was sent. Both outcomes record `launched:false` while
620
621
  retaining the real process/endpoint for inspection: this is never permission
621
622
  to allocate another harness.
@@ -636,11 +637,79 @@ No audit or metadata write is directed into the replacement home.
636
637
  A blocked result can follow a successfully submitted Enter. The accepted
637
638
  completion families cover exact installed-plugin and plugin-absent banners,
638
639
  plus source-derived billing-only variants of those empty frames (Max, Pro,
639
- Team and Enterprise). Other model, billing, effort or permission text can leave the pane active while
640
+ Team and Enterprise), and source-derived empty Max/bypass/medium frames.
641
+ Other model, billing, effort or permission text can leave the pane active while
640
642
  reporting a retained block. Inspect `answers` and the event receipts; never infer
641
643
  that nothing started or nothing was answered. See the
642
644
  [completion limitation](configuration.md#exact-home-launch-prompt-consent).
643
645
 
646
+ ### Private unmatched launch evidence
647
+
648
+ This section also covers mandatory structural-completion evidence.
649
+
650
+ After a submitted launch answer, an unmatched 110x35 frame is retained locally
651
+ at `<home>/.oats/launch-prompt-evidence/<sha256(startId)>/frame.txt`, with
652
+ `receipt.json` published last. This is the exact UTF-8 text already compared by
653
+ the controller, including whitespace and final newline, not a later recapture.
654
+ It is bounded to 64 KiB and 35 rows. There is no new input, observation, retry,
655
+ geometry change or readiness claim. Evidence is also mandatory when the post-answer
656
+ [structural completion rule](launch-prompt-completion.md) selects completion.
657
+ Previously qualified exact completion does not acquire that requirement. No
658
+ frame is saved before a submitted answer or for unrelated blocked reasons.
659
+
660
+ Find the invocation's existing `launch-prompt` event in the home's
661
+ `.oats-events.jsonl` (or the checked event rows in `launchPrompts.receipt`). Its
662
+ `data.startId` supplies the string to hash with SHA-256, UTF-8, yielding the
663
+ lowercase hexadecimal directory name. Match the event's home, socket, window,
664
+ pane, PID and `signatureDigest`; do not assume a top-level spawn-result `startId`.
665
+ Spawn and same-home start/restart use this same controller and event path.
666
+
667
+ New evidence directories are owner-only 0700 and files 0600. The writer checks
668
+ canonical directory identities, ownership, modes and symlinks, refuses ACLs on
669
+ evidence directories/files (including macOS ACLs hidden by xattrs), refuses an
670
+ existing invocation directory, and publishes bounded files without overwrite.
671
+ The private receipt binds the start, target, original home identity, dimensions,
672
+ frame inode, byte count and SHA-256. `retainedAt` is write time; `captureTime:null`
673
+ means no independent capture timestamp was recorded. It does not substitute an
674
+ event timestamp for capture time. Same-uid concurrent mutation between checks is
675
+ not excluded.
676
+
677
+ Private receipt version 2 uses `kind:"launch-prompt-frame"` and admits only
678
+ `outcome:"blocked", recognition:"unmatched"` or
679
+ `outcome:"completed", recognition:"structural"`. These fields describe the
680
+ recognition whose frame was saved, not the final public launch result: a later
681
+ audit or cleanup failure can still report `incomplete`. Historical version 1
682
+ receipts with `kind:"launch-prompt-unmatched"` describe blocked/unmatched
683
+ evidence. Readers must reject unknown versions, kinds or crossed pairs rather
684
+ than infer completion from an artifact's presence.
685
+
686
+ Before collecting a frame, verify its original `frameIdentity`, byte count,
687
+ digest and dimensions against the receipt, and bind the receipt's `startId`,
688
+ home and target to the checked launch event. A copied file has a different
689
+ inode: preserve the original identity as provenance instead of comparing the
690
+ copy's inode with it. Missing, partial, mismatched or failed-save evidence, or
691
+ an incomplete public launch result, must not be reported as a successful
692
+ completed observation. Collection never recaptures the pane or retries input.
693
+
694
+ Verify both files, receipt/identity/digest and the corresponding checked event
695
+ before treating the save as complete. A frame alone is partial evidence. Failed
696
+ publication, fsync, readback or temporary-file cleanup leaves the launch blocked
697
+ and adds only `; private launch evidence retention failed` to its diagnostic
698
+ reason. Already published files may remain, including a receipt if a later
699
+ verification fails; their presence alone is not a success claim. The submitted
700
+ answer is preserved. Existing event-audit failures still report `incomplete`.
701
+ A structural-completion save failure instead reports checked `incomplete` with
702
+ `private launch evidence retention failed`, preserving the submitted answer and
703
+ retained target. It never retries publication or input. An audit failure remains
704
+ incomplete even if private evidence was saved. No public result or event contains
705
+ the raw screen or private receipt.
706
+
707
+ The frame can contain private task or account text. It stays on this host with
708
+ no automatic transmission or expiry. Preserve needed evidence privately before
709
+ retirement; it follows the home's ordinary conservative recovery behavior, with
710
+ no integrity-neutral exclusion or new cleanup policy. A saved frame is evidence
711
+ for review, never automatic permission to activate a new matcher.
712
+
644
713
  Retiring a blocked or incomplete spawned home can preserve a recovery copy
645
714
  labelled `changed instance-home bytes`, even without authored changes: the
646
715
  kernel changed metadata from `launched:true` to `false` after its baseline.
@@ -13,8 +13,9 @@ reference pages ([workspaces](workspaces.md), [souls and instances](souls-and-in
13
13
  catalog (`package-catalog.json`). It has no runtime dependencies.
14
14
  - **`@awebai/oats-pi`** (`packages/pi/`): a thin pi adapter that exposes an
15
15
  instance's own resources. It registers no agent tools.
16
- - **`oats.framework`** (`oats-package/`): the `oats.core`, `oats.setup` and
17
- `oats.knowledge-theory` capabilities and the `knowledge-theory-expert` soul,
16
+ - **`oats.framework`** (`oats-package/`): the `oats.core`, `oats.setup`,
17
+ `oats.support` and `oats.knowledge-theory` capabilities and the
18
+ `knowledge-theory-expert` soul,
18
19
  released as a package under its own `oats-framework/v<version>` tags.
19
20
  `oats.core` is the one with executables: its spawn and launch hook and the
20
21
  Claude Code waiting emitter (`capabilities/oats-core/bin/`).
@@ -66,6 +67,8 @@ published to npm. Its developer docs are in
66
67
  | `capability-contract.mjs`, `provider-binding.mjs` | manifest validation (launch environment, hooks, `retirement`), the hook environment rules, the readiness wire |
67
68
  | `retire-output.mjs` | the lines `oats retire` prints for preserved work, one function for the local and the remote path |
68
69
  | `servers.mjs` | routing commands to a registered server |
70
+ | `launch-prompt-evidence.mjs` | private bounded storage of exact post-answer unmatched or structurally completed frames, original-home identity checks and receipt-last publication; no terminal operations or public raw text |
71
+ | `launch-prompt-completion.mjs` | pure, qualified post-answer completion reporting using a finite bottom input/footer table and 21 question markers; never an input matcher |
69
72
  | `launch-prompt-policy.mjs` | strict host-local exact-home aweb development-channel consent, canonical directory validation and config provenance |
70
73
  | `launch-prompts.mjs`, `launch-prompt-transport.mjs` | bounded launch-only controller and visible-screen transport; private process authority, injectable clock/transport, durable checked receipts and the fixture-gated exact aweb development-channel signature |
71
74
  | `harness-trust.mjs` | read-only native trust coverage and operator remedy for launches |
@@ -0,0 +1,70 @@
1
+ # Qualified post-answer completion
2
+
3
+ ## Boundary and accepted risk
4
+
5
+ Only after exactly one recorded **submitted** development-channel Enter, under the existing qualified executable/version/platform/argv, exact-home consent, original home identity, pane/PID and 110x35 checks, may structural recognition complete observation. It never authorizes input. Existing exact completion families remain available, subject to the marker precedence below. No structural recognition before an answer, after any terminal outcome, or after failed/uncertain input.
6
+
7
+ An unknown dialog can retain the composer and footer and evade this finite marker list. The contract explicitly accepts that completion-reporting risk. Completion closes input authority permanently; it does not establish absence of every dialog, idle state, account admission, channel readiness, or task success. The observed spinner is an in-progress task turn, not an idle prompt. No spinner word, glyph or timing grammar is introduced.
8
+
9
+ ## Source provenance
10
+
11
+ Read-only pinned Claude Code 2.1.289 darwin-arm64 binary SHA256 `03d66745e3bb69ec727d66023696f3820bc0a00a8a5ba725eb6706d0c67cbe69`. [the source manifest](../lib/launch-prompt-fixtures/structural-completion.json) records exact byte intervals and SHA256s for every reference below. No extracted module enters the repository. These are source-derived rules, not additional observed captures.
12
+
13
+ ## Qualified input/footer rule — non-flag footer layout
14
+
15
+ This qualifies the non-Jt footer layout (`tengu_copper_thistle` false) in the pinned source. The flag-enabled layout chooses a single hint instead of this list; it is not separately qualified. A remote flag change can alter the footer without changing the binary and reduce matches; only exact listed shapes match, never additional shapes.
16
+
17
+ Input is the same compared text object, exactly 35 newline-terminated rows at 110x35, bounded to 64 KiB. No whitespace normalization, cropping, ANSI stripping, case folding or transcript rewrite. Reject embedded CR, ESC and other C0 controls except LF. All rows after the footer must be empty. The footer must be the last nonempty row, with this immediately preceding contiguous region:
18
+
19
+ 1. Exactly 110 U+2500 horizontal-line characters.
20
+ 2. Exactly U+276F followed by U+00A0 (`❯` plus NBSP), and nothing else: the actual empty composer row, not a prompt-like substring in prose.
21
+ 3. Exactly 110 U+2500 horizontal-line characters.
22
+ 4. The finite footer below.
23
+
24
+ This accepts the region at any row where all four rows fit, not a fixed transcript height. It refuses nonempty/multiline editor drafts, shell or screen-reader selectors, annotated borders, wrapped footers, notifications between editor and footer, status rows below the footer, and footer-less states. It deliberately does not cover every normal renderer configuration.
25
+
26
+ Source: `prompt-selector` s0/qK renders Z.pointer+NBSP for the ordinary non-screen-reader input; `prompt-symbol` pins Z.pointer to U+276F. `round-border-glyphs` pins top/bottom to U+2500; `prompt-border` BA and `editor-layout` HWe give top/bottom borders with left/right borders absent. The existing raw API captures and the actual retained Max frame corroborate the exact three input rows at 110 columns. `footer-layout` YHe/Oie gives paddingX=2, row width=110, gap=1, right status marginLeft=auto. The sanitized regression frame's input region is rows 11–14; its home-only sanitized derivative is regression input, not an exact spinner matcher.
27
+
28
+ ### Modes (complete list)
29
+
30
+ | Native mode | Exact left label | Allowed cycle suffix |
31
+ |---|---|---|
32
+ | default | `⏸ manual mode on` | absent only |
33
+ | plan | `⏸ plan mode on` | absent or ` (shift+tab to cycle)` |
34
+ | acceptEdits | `⏵⏵ accept edits on` | absent or same cycle suffix |
35
+ | bypassPermissions | `⏵⏵ bypass permissions on` | absent or same cycle suffix |
36
+ | dontAsk | `⏵⏵ don't ask on` | absent or same cycle suffix |
37
+ | auto | `⏵⏵ auto mode on` | absent or same cycle suffix |
38
+
39
+ `mode-map` Sr/DL/bie/qQo enumerates all six; `mode-symbols` pins the symbols; `footer-mode` kx adds literal ` on` when bare is absent. `footer-branches` dHe calls kx without bare and gates the default-keybinding cycle hint on non-default mode and ks. Unknown modes, custom keybindings and bare labels are unsupported.
40
+
41
+ After the mode/cycle label, accept only these exact suffixes, each independently enumerated: empty; ` · ← for agents`; ` · esc to interrupt`; ` · esc to interrupt · ← for agents`. For **default mode only**, also ` · ? for shortcuts` or ` · ? for shortcuts · ← for agents`. Sources: `footer-branches` dHe/EFt (escape interrupt, shortcut hint, ordering), `agent-hint` Cze (left-arrow for agents). No return-to-team-lead, task counter, voice, custom status or arbitrary suffix. The normal hints are not negative markers anywhere in the visible frame; an unsupported footer remains unsupported for structural reasons rather than because it contains a question mark or Esc.
42
+
43
+ The footer is exactly two ASCII spaces + left text + ASCII space gap + one right label from this complete list: `○ low · /effort`, `◐ medium · /effort`, `● high · /effort`, `◉ xhigh · /effort`, `◈ max · /effort`. `effort` bW/y8 and `mode-symbols` define these glyphs and labels. The right label ends at column 108 (two columns right padding are absent in tmux text); gap is exactly `108 - 2 - leftDisplayWidth - rightDisplayWidth`, and at least one. All enumerated glyphs are single-column in the qualified captured geometry. No `effort:` prefix, ultracode, missing effort, altered punctuation, or other effort value. Existing medium-only exact families remain exact; the additional finite effort rules are source-qualified here, not prior capture qualification.
44
+
45
+ ## Qualified negative marker table
46
+
47
+ ONLY after exactly one recorded submitted answer, each listed string is matched **case-sensitively as exact contiguous text anywhere in the visible frame**, before either exact or structural positive completion. It can conservatively block a transcript quoting a known question. No regex/question-word heuristic, case folding, line joining, general `?`, general `Esc`, or exemption for an unknown question. Wrapped or customized unknown questions remain within the explicitly accepted risk. Input prompt matching itself is unchanged: the initial qualified channel question must still be answerable before a submitted answer exists.
48
+
49
+ | Exact forms (each independently tested) | Pinned source reference |
50
+ |---|---|
51
+ | `Accessing workspace:`; `Quick safety check:`; `Yes, I trust this folder`; `No, continue without these permissions`; `No, exit` | trust / TrustDialog |
52
+ | `WARNING: Loading development channels`; `I am using this for local development` | channel confirmation |
53
+ | `Managed settings require approval`; `Yes, I trust these settings`; `No, exit Claude Code` | managed-settings |
54
+ | `Do you want to use this API key?`; `Detected a custom API key in your environment` | api-key |
55
+ | `Select login method:` | login |
56
+ | `Do you want to proceed?` | permission-question / o5 default question |
57
+ | `Do you want to allow this connection?` | connection-question |
58
+ | `Waiting for permission…` | permission-wait / QVe |
59
+ | `Enter to confirm`; `Esc to cancel` | confirmation-guide and key-hint B renderer; gateway/selection-dialog corroborate cancel literal |
60
+ | `esc to cancel` | lower-cancel + key-hint B with lower keyCase |
61
+ | `Enter to select` | selection-dialog literal |
62
+ | `Esc to exit` | managed-settings + key-hint B |
63
+
64
+ The historical local-JSX confirmation is not claimed covered by this list; its marker composition was not separately verified and remains an example of the accepted unknown-dialog risk. Real #712 child screens with missing effort and update/focus notices below the footer remain unqualified; the retained A2 frame has the required effort and no trailing notices.
65
+
66
+ This is the whole v1 marker set, **not a claim that these are all possible native questions**. Generic custom title/confirm labels can evade it. `? for shortcuts` and `esc to interrupt` are explicitly normal UI, never markers. No case-insensitive expansion to `enter to confirm`, `esc to exit`, etc. without another qualified table update.
67
+
68
+
69
+ Private retention and receipt-v2 collection rules are documented in
70
+ [execution targets](execution-targets.md#private-unmatched-launch-evidence).
@@ -9,7 +9,7 @@ or workspace membership alone does not make a package official.
9
9
 
10
10
  | package | release | capabilities | package souls |
11
11
  |---|---|---|---|
12
- | `oats.framework` | `oats-framework/v1.6.1` (this repository) | `oats.core`, `oats.setup`, `oats.knowledge-theory` | `knowledge-theory-expert` |
12
+ | `oats.framework` | `oats-framework/v1.7.0` (this repository) | `oats.core`, `oats.setup`, `oats.knowledge-theory` | `knowledge-theory-expert` |
13
13
  | `oats.okf` | `v4.1.1` | `oats.okf` (knowledge), `oats.okf-harvest`, `oats.okf-maintenance` | `knowledge-harvester`, `knowledge-maintainer` |
14
14
  | `oats.aweb` | `v1.21.1` | `oats.aweb` (messaging) | |
15
15
  | `oats.engineering` | `v1.8.1` | `oats.engineering-expert`, `oats.developer`, `oats.code-review`, `oats.maintainer` | `code-reviewer` |
package/docs/packages.md CHANGED
@@ -51,7 +51,7 @@ packages:
51
51
  - **Bare version** (`v4.1.1`, `4.1.1`, `1.0.0-rc.1`): the id is looked up in
52
52
  the official catalog — `package-catalog.json` in the `oats` repo, or the file
53
53
  named by `OATS_PACKAGE_CATALOG` — which supplies the repo url, the tag
54
- convention (`v4.1.1` or `oats-framework/v1.6.1`) and the payload path. An id
54
+ convention (`v4.1.1` or `oats-framework/v1.7.0`) and the payload path. An id
55
55
  the catalog does not know is `E_PACKAGE_MISSING` ("use `git:<repo>@<ref>` for
56
56
  a package outside the catalog"). The catalog is the reviewed official list
57
57
  ([official-catalog.md](official-catalog.md)) and the only way a
@@ -74,7 +74,7 @@ members:
74
74
  - git:github.com/acme/agents
75
75
  - git:github.com/acme/platform
76
76
  packages:
77
- oats.framework: v1.6.1
77
+ oats.framework: v1.7.0
78
78
  oats.okf: v4.1.1
79
79
  oats.aweb: v1.21.1
80
80
  teams:
@@ -339,11 +339,11 @@ A soul that names one of the package's capabilities with
339
339
  "policy": "docs/official-catalog.md",
340
340
  "packages": {
341
341
  "oats.okf": { "url": "https://github.com/awebai/oats-okf.git", "ref": "v4.1.1", "path": "oats-package" },
342
- "oats.framework": { "url": "https://github.com/awebai/oats.git", "ref": "oats-framework/v1.6.1", "path": "oats-package" }
342
+ "oats.framework": { "url": "https://github.com/awebai/oats.git", "ref": "oats-framework/v1.7.0", "path": "oats-package" }
343
343
  }
344
344
  }
345
345
  ```
346
346
 
347
- `ref` carries the tag convention: a workspace's `oats.framework: v1.6.1`
348
- resolves to tag `oats-framework/v1.6.1`. Resolving through the catalog never
347
+ `ref` carries the tag convention: a workspace's `oats.framework: v1.7.0`
348
+ resolves to tag `oats-framework/v1.7.0`. Resolving through the catalog never
349
349
  advances a lock by itself: `oats sync` does, and says so.
@@ -0,0 +1,93 @@
1
+ # OATS 0.47.0
2
+
3
+ 2026-10-07
4
+
5
+ ## Changed
6
+
7
+ - **The Desktop accepts OATS CLIs `>=0.25.8 <0.48.0`.** The compatibility band
8
+ admits 0.47; existing feature gates remain unchanged.
9
+
10
+ - **Desktop: "Retire instance" says what will happen, and shows it happening.**
11
+ The roster row's menu item is **Retire instance…** (was "Remove instance…"),
12
+ and the confirmation reads **Retire <name>?** with a **Retire instance**
13
+ button. Stop shares the same dialog: **Stop <name>?**, **Stop session**
14
+ (**Stop sessions** for several).
15
+ - **Plain words.** **What will happen** lists the consequences from the
16
+ plan: the session stops, children stop first and are kept, a recovery copy
17
+ of uncommitted work is saved before the home folder is deleted, the
18
+ worktree is kept on its branch or deleted, and branches and pull requests
19
+ are not changed. **Current state** shows Running / Not running / Unknown,
20
+ the uncommitted work and the branch. The plan's time is shown as an age
21
+ ("Observed just now"), with no timestamps or kernel vocabulary.
22
+ - **Loading and progress.** The dialog opens with a skeleton and "Checking
23
+ what retiring <name> will do…". Changing **Also delete the worktree**
24
+ keeps the facts on screen until the new plan replaces them. Confirming
25
+ shows "Retiring <name>…" with a spinner, and says you can close the window
26
+ while retirement continues. Success ends with "<name> retired", a short
27
+ summary (where the worktree and the recovery copy are) and **Done**.
28
+ - **No unexplained disabled controls.** A control appears only in the
29
+ phase where it can act, and a waiting confirm button names why beside it.
30
+ **Cancel** becomes **Close** once something has been sent, since closing
31
+ never cancels it. **Refresh plan** is gone: **Check again** re-reads the
32
+ plan, and after a failure **Review again** does. An unconfirmed outcome
33
+ offers **Check again**, which resubmits the same confirmation.
34
+ - **The worktree option** appears only for an instance with its own
35
+ worktree. When checked, a warning says how many uncommitted changes the
36
+ folder holds and that the branch stays.
37
+ - **Row menu.** While one of its actions runs, the menu button stays
38
+ focusable and says "Waiting for <action> to finish". A failed action posts
39
+ a notification, with the error in Details, instead of an alert.
40
+ - The lifecycle messages are reworded in plain language, for example "OATS
41
+ didn't confirm the result. Check again before assuming nothing happened."
42
+ Their codes and the confirmation contract are unchanged. See
43
+ [Desktop Stop and Retire confirmations](../../packages/desktop/docs/desktop-lifecycle-confirmations.md#ui-states).
44
+
45
+ - **oats.framework 1.8.0: the new oats.support 1.0.0.** Instruction text
46
+ only; no kernel, schema or provider contract changes.
47
+ - **oats.support 1.0.0, a support desk role (new).** A resource-only
48
+ capability for a soul that anyone may reach. It takes in support
49
+ requests, turns each one into one ticket, hands each ticket to the
50
+ project's support maintainer, and keeps the requester informed.
51
+ Everything a requester sends is untrusted data. The desk works from a
52
+ fixed set of actions (acknowledge, ticket, hand off, relay, a private
53
+ security route, ask its human). It never merges, approves, closes or
54
+ pushes, and never executes what it receives. A requester's words never
55
+ reach another agent as instructions: they stay quoted and fenced in the
56
+ ticket. When the support maintainer is in no roster, the desk starts the
57
+ stopped one on its machine or spawns one from its soul's recipe, at most
58
+ once per spawn interval, with a brief that holds ticket references only
59
+ and grants no authority. A support maintainer that is listed but silent
60
+ goes to the desk's human instead. Skills: `/support-intake`,
61
+ `/support-safety`, `/support-ticket`, `/support-handoff`,
62
+ `/support-relay`. The capability is generic: the tracker, the support
63
+ maintainer, the private security route and the limits are the soul's.
64
+ Its kernel floor is `>=0.40.0`, for `oats instance attention`. A
65
+ workspace gets it by moving its pin to `oats.framework: v1.8.0` and
66
+ running `oats sync`. The repository's package checks
67
+ (`scripts/check-knowledge-theory-package.mjs`, the clean-room smoke)
68
+ pin the distribution's capability list, so they list it too. They ship
69
+ nothing.
70
+
71
+ ## Fixed
72
+
73
+ - **Post-answer Claude completion and private evidence:** after the sole
74
+ submitted channel Enter, the pinned Claude 2.1.289 darwin-arm64 110x35 controller
75
+ can recognize a bounded bottom composer/effort-footer region with six native
76
+ mode labels, five effort labels and finite hints. Twenty-one exact known
77
+ question markers block before either exact or structural completion. Normal
78
+ shortcut/interrupt hints do not block. The non-Jt footer layout is qualified;
79
+ flag drift, missing effort and trailing notices can still refuse completion.
80
+ **An unknown dialog retaining the region can be misclassified completed.**
81
+ This accepted reporting risk cannot authorize another key or prove idle,
82
+ account or channel readiness. Existing Enter and consent/ownership guards are
83
+ unchanged. A home-only sanitized derivative of an actual in-progress Max
84
+ comparison is the regression fixture; there is no exact spinner matcher or
85
+ new live proof of the revised classifier. Also adds the source-qualified
86
+ empty Max/bypass/medium families.
87
+ Exact compared bytes are retained privately for structural completion and
88
+ post-answer unmatched blocks. Receipt v2 distinguishes those two outcomes;
89
+ historical v1 remains readable. A structural save failure reports checked
90
+ incomplete with answer/target preserved. No raw text enters public output.
91
+ See [completion scope](../launch-prompt-completion.md) and
92
+ [private evidence](../execution-targets.md#private-unmatched-launch-evidence).
93
+ This does not close #754 or establish general onboarding success.
@@ -102,7 +102,9 @@ souls) is ordinary capability content: **`oats.core`** (package
102
102
  `oats.framework`). Workspaces give it to every soul through
103
103
  `defaults.capabilities: { oats.core: { from: package } }`; a soul may say
104
104
  `oats.core: off`. **`oats.setup`** (same package) carries the whole-architecture
105
- knowledge an onboarding expert needs. Neither is kernel magic; the kernel still
105
+ knowledge an onboarding expert needs, and **`oats.support`** (same package)
106
+ is the support desk role: a publicly reachable soul that tickets requests and
107
+ hands them to a support maintainer. None of them is kernel magic; the kernel still
106
108
  composes its own instance-boundary and work-mode briefings, and the "You run
107
109
  on OATS" briefing is `oats.core`'s inject: a soul without `oats.core` gets no
108
110
  OATS operating instructions, and `oats doctor --soul` says so. Its capability
@@ -54,7 +54,7 @@ members: # repo refs, NO @revision (E_WORKSPAC
54
54
  - git:github.com/acme/tools # a member that ALSO publishes a package (see below)
55
55
 
56
56
  packages: # the ONLY versioned things
57
- oats.framework: v1.6.1 # bare version → resolves through the official catalog
57
+ oats.framework: v1.7.0 # bare version → resolves through the official catalog
58
58
  oats.okf: v4.1.1
59
59
  acme.tools: git:github.com/acme/tools@v0.4.0 # outside the catalog → git:<repo>@<tag|OID>; still a package
60
60
 
package/lib/core.mjs CHANGED
@@ -29,6 +29,7 @@
29
29
  */
30
30
  import { effectiveLaunchPromptPolicy } from "./launch-prompt-policy.mjs";
31
31
  import { createLaunchPromptController } from "./launch-prompts.mjs";
32
+ import { retainLaunchPromptFrame } from "./launch-prompt-evidence.mjs";
32
33
  import { qualifyLaunchPromptFixtures } from "./launch-prompt-fixtures.mjs";
33
34
  import { LAUNCH_PANE_FORMAT, launchPaneTarget, launchPromptTransport } from "./launch-prompt-transport.mjs";
34
35
  import { execFileSync, execSync, spawn as spawnProcess, spawnSync } from "node:child_process";
@@ -646,7 +647,7 @@ function promptController(home, startId, harness, policy, command, io, expectedH
646
647
  const described = describeLaunchCommand(command);
647
648
  const qualification = qualifyLaunchPromptFixtures({ home: realPathOrNearest(home), harness, executablePath: described.executable, argv: described.argv });
648
649
  return createLaunchPromptController({ home: realPathOrNearest(home), startId, ...qualification, policy, expectedHomeIdentity,
649
- transport: launchPromptTransport(io), audit: data => appendEvent(home, { kind: "launch-prompt", data }),
650
+ transport: launchPromptTransport(io), audit: data => appendEvent(home, { kind: "launch-prompt", data }), retainFrame: retainLaunchPromptFrame,
650
651
  ...(io?.now ? { now: io.now } : {}), ...(io?.sleep ? { sleep: io.sleep } : {}) });
651
652
  }
652
653
  function retainedPromptError(home, instance, target, parentLineageCommitted, launchPrompts) {
@@ -0,0 +1,67 @@
1
+ /** Completion reporting only, never input authority. Source-qualified Claude
2
+ * 2.1.289 non-Jt footer layout; the caller must enforce qualification and one
3
+ * submitted answer. Unknown dialogs retaining this region can evade the finite
4
+ * markers. That reporting risk is accepted; no subsequent key may be sent.
5
+ * See launch-prompt-fixtures/structural-completion.json and README.md. */
6
+ export const CLAUDE_QUESTION_MARKERS = Object.freeze([
7
+ 'Accessing workspace:',
8
+ 'Quick safety check:',
9
+ 'Yes, I trust this folder',
10
+ 'No, continue without these permissions',
11
+ 'No, exit',
12
+ 'WARNING: Loading development channels',
13
+ 'I am using this for local development',
14
+ 'Managed settings require approval',
15
+ 'Yes, I trust these settings',
16
+ 'No, exit Claude Code',
17
+ 'Do you want to use this API key?',
18
+ 'Detected a custom API key in your environment',
19
+ 'Select login method:',
20
+ 'Do you want to proceed?',
21
+ 'Do you want to allow this connection?',
22
+ 'Waiting for permission…',
23
+ 'Enter to confirm',
24
+ 'Esc to cancel',
25
+ 'esc to cancel',
26
+ 'Enter to select',
27
+ 'Esc to exit',
28
+ ]);
29
+
30
+ const modes = [
31
+ ['default', '⏸ manual mode on'], ['plan', '⏸ plan mode on'],
32
+ ['acceptEdits', '⏵⏵ accept edits on'], ['bypassPermissions', '⏵⏵ bypass permissions on'],
33
+ ['dontAsk', "⏵⏵ don't ask on"], ['auto', '⏵⏵ auto mode on'],
34
+ ];
35
+ const efforts = ['○ low · /effort', '◐ medium · /effort', '● high · /effort', '◉ xhigh · /effort', '◈ max · /effort'];
36
+ const hints = ['', ' · ← for agents', ' · esc to interrupt', ' · esc to interrupt · ← for agents'];
37
+ const shortcuts = [' · ? for shortcuts', ' · ? for shortcuts · ← for agents'];
38
+ const footers = new Set();
39
+ for (const [mode, label] of modes) {
40
+ for (const cycle of mode === 'default' ? [''] : ['', ' (shift+tab to cycle)']) {
41
+ for (const hint of mode === 'default' ? [...hints, ...shortcuts] : hints) {
42
+ for (const right of efforts) {
43
+ const left = label + cycle + hint;
44
+ // Every enumerated character is single-column in the qualified layout.
45
+ const gap = 108 - 2 - left.length - right.length;
46
+ if (gap >= 1) footers.add(' ' + left + ' '.repeat(gap) + right);
47
+ }
48
+ }
49
+ }
50
+ }
51
+ const border = '─'.repeat(110);
52
+
53
+ export function classifyClaudeLaunchCompletion(screen) {
54
+ if (screen?.width !== 110 || screen?.height !== 35 || typeof screen.text !== 'string'
55
+ || Buffer.byteLength(screen.text, 'utf8') > 64 * 1024) return null;
56
+ // No exemptions for unknown questions. Ordinary shortcut/interrupt hints are
57
+ // simply not markers; do not broaden this into a generic '?' or Esc rule.
58
+ if (CLAUDE_QUESTION_MARKERS.some(marker => screen.text.includes(marker))) return 'blocked';
59
+ if (/[\x00-\x09\x0b-\x1f]/.test(screen.text)) return null;
60
+ const rows = screen.text.split('\n');
61
+ if (rows.length !== 36 || rows.pop() !== '') return null;
62
+ let footer = rows.length - 1;
63
+ while (footer >= 0 && rows[footer] === '') footer--;
64
+ if (footer < 3 || !footers.has(rows[footer]) || rows[footer - 1] !== border
65
+ || rows[footer - 2] !== '❯\u00a0' || rows[footer - 3] !== border) return null;
66
+ return 'completed';
67
+ }
@@ -0,0 +1,159 @@
1
+ /** Private, bounded evidence from an already compared launch frame. No terminal
2
+ * operations or public raw-text diagnostics. A receipt is published last; a
3
+ * partial directory is retained on failure, never presented as a complete save. */
4
+ import * as fs from 'node:fs';
5
+ import { join, dirname, isAbsolute } from 'node:path';
6
+ import { createHash, randomUUID } from 'node:crypto';
7
+ import { execFileSync } from 'node:child_process';
8
+
9
+ const sha256 = bytes => createHash('sha256').update(bytes).digest('hex');
10
+ const same = (a, b) => a && b && a.dev === b.dev && a.ino === b.ino;
11
+ const fail = () => { throw new Error('private launch evidence unavailable'); };
12
+ const uid = () => process.getuid?.();
13
+ const identity = s => ({ dev: s.dev, ino: s.ino });
14
+
15
+ /** Decode only the private receipt's recognition tag. This is not evidence
16
+ * verification: collection must still bind bytes/identities and checked public
17
+ * outcome. Historical receipts remain unchanged on disk. */
18
+ export function launchPromptEvidenceRecognition(receipt) {
19
+ if (receipt?.version === 1 && receipt.kind === 'launch-prompt-unmatched'
20
+ && !Object.hasOwn(receipt, 'outcome') && !Object.hasOwn(receipt, 'recognition')) {
21
+ return { outcome: 'blocked', recognition: 'unmatched' };
22
+ }
23
+ if (receipt?.version === 2 && receipt.kind === 'launch-prompt-frame'
24
+ && ((receipt.outcome === 'blocked' && receipt.recognition === 'unmatched')
25
+ || (receipt.outcome === 'completed' && receipt.recognition === 'structural'))) {
26
+ return { outcome: receipt.outcome, recognition: receipt.recognition };
27
+ }
28
+ return null;
29
+ }
30
+
31
+ // Mode bits alone do not rule out an inherited macOS ACL (including one
32
+ // hidden behind the '@' xattr indicator). Never write private bytes first and
33
+ // then discover that another principal can read the candidate.
34
+ function privateAccess(path) {
35
+ const mac = process.platform === 'darwin';
36
+ const output = execFileSync('/bin/ls', mac ? ['-lde', path] : ['-ld', '--', path],
37
+ { encoding: 'utf8', stdio: ['ignore', 'pipe', 'pipe'], timeout: 1000, maxBuffer: 16384, env: { ...process.env, LC_ALL: 'C' } });
38
+ const mode = output.split(/\s/, 1)[0];
39
+ if (mac) {
40
+ if (!/^[d-][-rwxStTs]{9}@?$/.test(mode) || output.trimEnd().split('\n').length !== 1) fail();
41
+ } else if (!/^[d-][-rwxStTs]{9}\.?$/.test(mode)) fail();
42
+ }
43
+
44
+ export function retainLaunchPromptFrame({ home, homeIdentity, startId, target, screen,
45
+ outcome = 'blocked', recognition = 'unmatched' }) {
46
+ const directories = new Map(), temps = new Map();
47
+ let ok = false;
48
+ function directory(path, privateMode = false) {
49
+ const s = fs.lstatSync(path);
50
+ if (!s.isDirectory() || s.uid !== uid() || (s.mode & 0o7022)
51
+ || (privateMode && (s.mode & 0o777) !== 0o700)
52
+ || fs.realpathSync.native(path) !== path) fail();
53
+ return s;
54
+ }
55
+ function check() {
56
+ if (!same(directory(home), homeIdentity)) fail();
57
+ for (const [path, saved] of directories) if (!same(directory(path, saved.privateMode), saved)) fail();
58
+ }
59
+ function ensure(path, privateMode) {
60
+ check();
61
+ try { fs.mkdirSync(path, { mode: 0o700 }); }
62
+ catch (error) { if (error.code !== 'EEXIST') throw error; }
63
+ const s = directory(path, privateMode);
64
+ if (privateMode) privateAccess(path);
65
+ directories.set(path, { ...identity(s), privateMode });
66
+ check();
67
+ syncDirectory(dirname(path));
68
+ }
69
+ function syncDirectory(path) {
70
+ check();
71
+ const fd = fs.openSync(path, fs.constants.O_RDONLY | (fs.constants.O_NOFOLLOW ?? 0) | (fs.constants.O_NONBLOCK ?? 0));
72
+ try {
73
+ if (!same(fs.fstatSync(fd), directories.get(path))) fail();
74
+ fs.fsyncSync(fd);
75
+ } finally { fs.closeSync(fd); }
76
+ check();
77
+ }
78
+ function verify(path, saved, bytes) {
79
+ check();
80
+ privateAccess(path);
81
+ const fd = fs.openSync(path, fs.constants.O_RDONLY | (fs.constants.O_NOFOLLOW ?? 0) | (fs.constants.O_NONBLOCK ?? 0));
82
+ try {
83
+ const s = fs.fstatSync(fd);
84
+ if (!s.isFile() || !same(s, saved) || s.nlink !== 1 || s.uid !== uid() || (s.mode & 0o7777) !== 0o600 || s.size !== bytes.length) fail();
85
+ // Read at most the expected bytes plus one: concurrent growth must not
86
+ // turn verification of a bounded frame into an unbounded file read.
87
+ const observed = Buffer.alloc(bytes.length + 1);
88
+ let length = 0;
89
+ while (length < observed.length) {
90
+ const count = fs.readSync(fd, observed, length, observed.length - length, length);
91
+ if (!count) break;
92
+ length += count;
93
+ }
94
+ if (length !== bytes.length || !observed.subarray(0, length).equals(bytes)) fail();
95
+ const after = fs.lstatSync(path);
96
+ if (!after.isFile() || !same(after, saved) || after.nlink !== 1 || after.uid !== uid()
97
+ || (after.mode & 0o7777) !== 0o600 || after.size !== bytes.length) fail();
98
+ } finally { fs.closeSync(fd); }
99
+ check();
100
+ }
101
+ function publish(dir, name, bytes) {
102
+ check();
103
+ const temp = join(dir, `.${randomUUID()}.tmp`), path = join(dir, name);
104
+ const fd = fs.openSync(temp, fs.constants.O_WRONLY | fs.constants.O_CREAT | fs.constants.O_EXCL | (fs.constants.O_NOFOLLOW ?? 0), 0o600);
105
+ let saved;
106
+ try {
107
+ saved = identity(fs.fstatSync(fd)); temps.set(temp, saved);
108
+ privateAccess(temp);
109
+ check();
110
+ if (!same(fs.lstatSync(temp), saved)) fail();
111
+ fs.writeFileSync(fd, bytes); fs.fsyncSync(fd);
112
+ } finally { fs.closeSync(fd); }
113
+ check();
114
+ // link is atomic and refuses ANY existing destination. Unlike rename it
115
+ // cannot overwrite a competing receipt or file between checks.
116
+ fs.linkSync(temp, path);
117
+ if (!same(fs.lstatSync(temp), saved)) fail();
118
+ fs.unlinkSync(temp); temps.delete(temp);
119
+ syncDirectory(dir);
120
+ verify(path, saved, bytes);
121
+ return saved;
122
+ }
123
+ try {
124
+ // This private receipt records the recognition being saved, not the final
125
+ // public launch result: a subsequent audit/cleanup failure can be incomplete.
126
+ if (!launchPromptEvidenceRecognition({ version: 2, kind: 'launch-prompt-frame', outcome, recognition })) fail();
127
+ if (typeof home !== 'string' || !isAbsolute(home) || !homeIdentity
128
+ || typeof startId !== 'string' || !startId.length || startId.length > 256 || startId.includes('\0')
129
+ || !target || !isAbsolute(target.socket || '') || !/^@\d+$/.test(target.windowId) || !/^%\d+$/.test(target.paneId) || !/^[1-9]\d*$/.test(String(target.pid))
130
+ || screen?.width !== 110 || screen?.height !== 35 || typeof screen.text !== 'string'
131
+ || !screen.text.endsWith('\n') || screen.text.split('\n').length !== 36) fail();
132
+ const bytes = Buffer.from(screen.text, 'utf8');
133
+ if (bytes.length > 64 * 1024) fail();
134
+ check();
135
+ directories.set(home, { ...identity(homeIdentity), privateMode: false });
136
+ const oats = join(home, '.oats'), root = join(oats, 'launch-prompt-evidence');
137
+ ensure(oats, false); ensure(root, true);
138
+ const dir = join(root, sha256(startId));
139
+ check(); fs.mkdirSync(dir, { mode: 0o700 }); // no overwrite/retry of an invocation
140
+ directories.set(dir, { ...identity(directory(dir, true)), privateMode: true });
141
+ privateAccess(dir);
142
+ syncDirectory(root);
143
+ const frameIdentity = publish(dir, 'frame.txt', bytes);
144
+ const receipt = Buffer.from(JSON.stringify({ version: 2, kind: 'launch-prompt-frame', outcome, recognition, home, homeIdentity: identity(homeIdentity), startId,
145
+ target: { socket: target.socket, windowId: target.windowId, paneId: target.paneId, pid: String(target.pid) },
146
+ width: 110, height: 35, bytes: bytes.length, signatureDigest: sha256(bytes), frame: 'frame.txt', frameIdentity,
147
+ source: 'controller-compared-frame', captureTime: null, retainedAt: new Date().toISOString() }) + '\n');
148
+ publish(dir, 'receipt.json', receipt);
149
+ verify(join(dir, 'frame.txt'), frameIdentity, bytes);
150
+ ok = true;
151
+ } catch { /* Do not expose raw text, OS errors, or a fabricated saved receipt. */ }
152
+ finally {
153
+ for (const [path, saved] of temps) {
154
+ try { check(); if (!same(fs.lstatSync(path), saved)) fail(); fs.unlinkSync(path); }
155
+ catch { ok = false; }
156
+ }
157
+ }
158
+ return { ok };
159
+ }
@@ -28,7 +28,7 @@ blocked outcome and sends no key. Captured folder-trust screens live only in
28
28
 
29
29
  Normal-output completion has a separate accepted finite policy: all surrounding
30
30
  bytes must equal one of the captured normal banners, except for its bounded
31
- directory line and the source-derived billing atoms described below. Candidates are the exact home, a home-directory-relative `~` spelling,
31
+ directory line and the source-derived header/footer variants described below. Candidates are the exact home, a home-directory-relative `~` spelling,
32
32
  and `~/…/` followed by nonempty whole-component suffixes within the user home.
33
33
  Every complete line must fit 110 columns. Only conservative ASCII path components
34
34
  without traversal or empty segments qualify. These variants can only terminate
@@ -45,8 +45,8 @@ network and real-home access, seeded onboarding/trust/feature flag, one Enter,
45
45
  stable bytes at 200ms/1s/3s, and verified private-server cleanup. The prompt and
46
46
  blank transition were byte-identical to the earlier captures. No configuration,
47
47
  credentials or extracted source is included. No task or real-account/channel
48
- connection was tested. Apart from the billing atoms below, other
49
- model/billing/effort/permission variants remain unqualified; a submitted answer can be followed by a retained blocked result.
48
+ connection was tested. Apart from the source-derived variants below, other
49
+ model/billing/effort/permission variants are not exact-frame candidates; a submitted answer can be followed by a retained blocked result.
50
50
 
51
51
  ## Source-derived subscription billing variants
52
52
 
@@ -76,9 +76,77 @@ proprietary extracted module is included.
76
76
  The reported Max journey is different evidence: its later retained pane has
77
77
  SHA256 `9eb587ab3ac7f97b746169619e02e3c8dd7df2c9069ff4368b6f0858c0964c68`,
78
78
  1597 bytes and 51 newline rows. It shows task activity and bypass-permissions
79
- mode. Captured at 2026-10-07T14:31:54.028296Z, it follows the submitted Enter at
79
+ mode. Its file mtime is 2026-10-07T14:31:54.028296Z (not an independently recorded
80
+ capture time); it follows the submitted Enter at
80
81
  14:31:45.365Z and block at 14:31:45.476Z. Its geometry, executable digest and
81
82
  plugin version were not independently recorded; no earlier raw frame exists.
82
83
  It is not padded, cropped or claimed as a qualified 110x35 first frame. See the
83
84
  [published operator evidence](https://github.com/awebai/oats/issues/754#issuecomment-6040215059).
84
85
  The billing-only extension does not recognize that later task-filled screen.
86
+
87
+ ## Source-derived empty Max/bypass/medium variants
88
+
89
+ **SOURCE-DERIVED / SYNTHETIC:** only the two existing empty banners with
90
+ `Claude Max` additionally qualify the bypass-permissions footer. The pinned
91
+ source ranges and hashes, exact changed row indices, sample home and complete
92
+ candidate digests are in
93
+ [the derivation manifest](claude-2.1.289-darwin-arm64/max-bypass-source.json).
94
+ No proprietary extracted modules or private task text are included.
95
+
96
+ The permission-mode table uses the same `⏵⏵` symbol for auto and bypass; only
97
+ colour differs, which plain captures omit. `kx` renders that symbol, the mode
98
+ indicator and ` on`; both modes take `dHe`'s same non-default branch with the
99
+ same hint. Thus the left text changes from `auto mode on` to
100
+ `bypass permissions on`, adding nine columns. The footer container has width
101
+ 110, horizontal padding 2 and a right status child with `marginLeft:auto`.
102
+ The alignment gap shrinks from 37 to 28 spaces; the effort glyph remains at
103
+ zero-based column 90 and the complete row remains 108 columns. The independent
104
+ `y8` effort rendering stays exactly `◐ medium · /effort`. All other bytes are
105
+ unchanged, except the already qualified header atom and finite home line.
106
+
107
+ The source supports this bounded transformation, not arbitrary renderer states.
108
+ Within these exact families, the displayed medium status is mandatory; other
109
+ content does not match an exact candidate. The separate post-answer structural
110
+ policy below can recognize additional bounded states. The #712 observed child screens corroborate the 62-column left footer
111
+ text, but have task/update rows and no effort status there; they are not these
112
+ fixtures. Full-frame comparison and all Enter guards remain unchanged.
113
+
114
+ Four later private operator captures motivated this work. Their SHA-256 values
115
+ are retained here as provenance only:
116
+
117
+ | Evidence | Bytes / newline rows | SHA-256 |
118
+ | --- | --- | --- |
119
+ | Journey start | 1597 / 51 | `9eb587ab3ac7f97b746169619e02e3c8dd7df2c9069ff4368b6f0858c0964c68` |
120
+ | Journey restart | 2449 / 51 | `c21cc6ac90487a18828ea2375243c642df2d6959b50a339a187132ec26dc213a` |
121
+ | Pre-rehearsal spawn | 3604 / 87 | `368170ecf1155e45f92a9c394972bf5a2ff0d56d2e03501f3d88615d0d2fab67` |
122
+ | Pre-rehearsal restart | 3180 / 87 | `88352f740e3e8f51f3613a057d7fe8aa41ae26efe20fb0ce24c7ac025fa70d9c` |
123
+
124
+ All follow controller geometry restoration; actual later-capture geometry,
125
+ executable hash and plugin version were not independently recorded. File mtime
126
+ is not capture time, and restoration dimensions are not capture dimensions.
127
+ No first post-answer frame survives. None is cropped, reflowed or published as
128
+ a raw 110x35 replay. These four captures remain provenance only, not qualified structural input.
129
+ The separate controller-compared regression and structural policy below do not
130
+ turn them into qualified captures or establish general Max success.
131
+
132
+ ## Post-answer structural completion
133
+
134
+ The [complete table](../../docs/launch-prompt-completion.md) defines the six modes,
135
+ five efforts, finite hints, exact region geometry and 21 marker forms.
136
+ [Source provenance](structural-completion.json) pins all 25 byte ranges against
137
+ the same executable hash. This is the non-Jt footer layout; remote feature flags
138
+ can change shape without binary drift. Unknown dialogs retaining a recognized
139
+ composer/footer can evade the marker set: the accepted risk concerns reporting
140
+ only, never another input action or a readiness claim.
141
+
142
+ The observed regression is a SANITIZED DERIVATIVE of the actual compared
143
+ 110x35 frame retained by the earlier controller after its one Enter. Only
144
+ zero-based home row 2 becomes `/fixture/home`; the manifest records separate
145
+ original/derivative byte counts and hashes. Outside that row, independent review
146
+ found only native constants and the launcher's literal `@TASK.md` reference, no
147
+ task body, account-personal or routing data. Four additional rows distinguish
148
+ this captured in-progress state from the empty source-derived family. Its
149
+ spinner verb is chosen per turn (or configured), its glyph animates and later
150
+ suffixes change. No exact spinner matcher or recurrence claim follows. The
151
+ empty footer derivation remains separately qualified; its arithmetic was not
152
+ disproved by observing a different startup state.
@@ -0,0 +1,88 @@
1
+ {
2
+ "binarySha256": "03d66745e3bb69ec727d66023696f3820bc0a00a8a5ba725eb6706d0c67cbe69",
3
+ "method": "read-only byte slices, binary not executed; no extracted modules copied",
4
+ "ranges": [
5
+ {
6
+ "description": "permission enum, indicator/symbol/color selectors",
7
+ "startInclusive": 179784383,
8
+ "endExclusive": 179785973,
9
+ "sha256": "290bd33df357cf05d275997cbe491bcb1689947829d9f106d6b1b9d2235384aa"
10
+ },
11
+ {
12
+ "description": "mode footer component kx",
13
+ "startInclusive": 208947546,
14
+ "endExclusive": 208948093,
15
+ "sha256": "bb9a808a262604016a60b52371acabdf71eb929a3d22a1d3b060e48cec8389cd"
16
+ },
17
+ {
18
+ "description": "footer state layout dHe",
19
+ "startInclusive": 209072435,
20
+ "endExclusive": 209082284,
21
+ "sha256": "3c52ece98f866e45e9dcc804616d2e601a58bd0d4b637ef3745fadc7ac0cae9d"
22
+ },
23
+ {
24
+ "description": "footer parent and right-aligned status",
25
+ "startInclusive": 209093603,
26
+ "endExclusive": 209099807,
27
+ "sha256": "5d9c5a87fdcbf304bce467c1569f7ddb6e10267929ba68184127a2adf7c6dda2"
28
+ },
29
+ {
30
+ "description": "effort label formatting",
31
+ "startInclusive": 208496116,
32
+ "endExclusive": 208496448,
33
+ "sha256": "840a01dd29dc8a6ccb0bdd5d4d4b4e93ac44711552725afa5bdbe278f31b846b"
34
+ }
35
+ ],
36
+ "candidates": [
37
+ {
38
+ "sourceCapture": "frame-07-after-channel-200ms.txt",
39
+ "sourceSha256": "1b30359fcb9181d02faf10ed0780812079222cf6c54b60ed468ce037c1ff37e7",
40
+ "candidateSha256": "e223df1791380aa2632bec04e584a821684991d1ce60a70426036dbcf5105c60",
41
+ "candidateBytes": 1194,
42
+ "rows": 35,
43
+ "geometry": {
44
+ "width": 110,
45
+ "height": 35
46
+ },
47
+ "changedRowIndices": [
48
+ 1,
49
+ 2,
50
+ 11
51
+ ],
52
+ "footer": {
53
+ "index": 11,
54
+ "oldGapSpaces": 37,
55
+ "newGapSpaces": 28,
56
+ "effortColumnZeroBased": 90,
57
+ "text": " \u23f5\u23f5 bypass permissions on (shift+tab to cycle) \u00b7 \u2190 for agents \u25d0 medium \u00b7 /effort"
58
+ },
59
+ "status": "SOURCE-DERIVED / SYNTHETIC, not an observed capture"
60
+ },
61
+ {
62
+ "sourceCapture": "frame-08-installed-after-channel-200ms.txt",
63
+ "sourceSha256": "f27d73189f2b1ba23899d99ba1596545f29d1d2494cb94b12ab314cf42426bef",
64
+ "candidateSha256": "0c83c1ccead19bfcd9bd6947a284948b30c551eae339258f66b6fe59b063c2e1",
65
+ "candidateBytes": 1128,
66
+ "rows": 35,
67
+ "geometry": {
68
+ "width": 110,
69
+ "height": 35
70
+ },
71
+ "changedRowIndices": [
72
+ 1,
73
+ 2,
74
+ 10
75
+ ],
76
+ "footer": {
77
+ "index": 10,
78
+ "oldGapSpaces": 37,
79
+ "newGapSpaces": 28,
80
+ "effortColumnZeroBased": 90,
81
+ "text": " \u23f5\u23f5 bypass permissions on (shift+tab to cycle) \u00b7 \u2190 for agents \u25d0 medium \u00b7 /effort"
82
+ },
83
+ "status": "SOURCE-DERIVED / SYNTHETIC, not an observed capture"
84
+ }
85
+ ],
86
+ "qualification": "SOURCE-DERIVED / SYNTHETIC complete empty Max/bypass/medium frames; not observed subscription captures",
87
+ "sampleHome": "/fixture/home"
88
+ }
@@ -0,0 +1,205 @@
1
+ {
2
+ "binarySha256": "03d66745e3bb69ec727d66023696f3820bc0a00a8a5ba725eb6706d0c67cbe69",
3
+ "version": "2.1.289",
4
+ "platform": "darwin-arm64",
5
+ "ranges": [
6
+ {
7
+ "id": "mode-symbols",
8
+ "start": 179782256,
9
+ "end": 179784383,
10
+ "sha256": "cc59173a00e4391864795bd39fa4da775cff19da9d858c3723b8878b7e4b1933"
11
+ },
12
+ {
13
+ "id": "mode-map",
14
+ "start": 179784383,
15
+ "end": 179785973,
16
+ "sha256": "290bd33df357cf05d275997cbe491bcb1689947829d9f106d6b1b9d2235384aa"
17
+ },
18
+ {
19
+ "id": "key-hint",
20
+ "start": 197498600,
21
+ "end": 197502800,
22
+ "sha256": "d666e7d25425520ac1b2e90f2e14e797b343d41df5f857ae645ec8f8c0f82103"
23
+ },
24
+ {
25
+ "id": "prompt-symbol",
26
+ "start": 187133430,
27
+ "end": 187133650,
28
+ "sha256": "29b53ed2eb87a241383e548d981fa1296bf72a386b46166e850b104d6201b220"
29
+ },
30
+ {
31
+ "id": "prompt-selector",
32
+ "start": 209099897,
33
+ "end": 209100980,
34
+ "sha256": "c8309eedb10e283a5a41261204c9dd8212b21bc50c9bbe2cf8d49192120b981a"
35
+ },
36
+ {
37
+ "id": "prompt-border",
38
+ "start": 209124438,
39
+ "end": 209125440,
40
+ "sha256": "61ba81594d61dc978f8e244c831465fc8fa4d0047c59a95136af9f85d41074e6"
41
+ },
42
+ {
43
+ "id": "editor-layout",
44
+ "start": 209158000,
45
+ "end": 209161530,
46
+ "sha256": "6f99e5bb96a73aa69fa977280c375e5829680041911e83aa634c28934e802b5f"
47
+ },
48
+ {
49
+ "id": "footer-mode",
50
+ "start": 208947546,
51
+ "end": 208948093,
52
+ "sha256": "bb9a808a262604016a60b52371acabdf71eb929a3d22a1d3b060e48cec8389cd"
53
+ },
54
+ {
55
+ "id": "footer-branches",
56
+ "start": 209072435,
57
+ "end": 209082570,
58
+ "sha256": "ea8c896b8d6225c76beddd9be6bd5d9c4d18ccefb15e35b79c725f2eafe45fe0"
59
+ },
60
+ {
61
+ "id": "footer-layout",
62
+ "start": 209093603,
63
+ "end": 209099807,
64
+ "sha256": "5d9c5a87fdcbf304bce467c1569f7ddb6e10267929ba68184127a2adf7c6dda2"
65
+ },
66
+ {
67
+ "id": "agent-hint",
68
+ "start": 209052245,
69
+ "end": 209055035,
70
+ "sha256": "4e528a20ee2cb4b931839f8eb72e33aada6940193e71c888c284e267fd3a974d"
71
+ },
72
+ {
73
+ "id": "effort",
74
+ "start": 208495700,
75
+ "end": 208496448,
76
+ "sha256": "ca7197f1491a673124e7ecc57e2c93198df1a8bdd4aa1bce9e1e0324ac5117a1"
77
+ },
78
+ {
79
+ "id": "trust",
80
+ "start": 213600100,
81
+ "end": 213603100,
82
+ "sha256": "a45e19812ac9f3d6002c2f9f6e4de3a2b156d83626712edaba016a9863d4f11a"
83
+ },
84
+ {
85
+ "id": "channel",
86
+ "start": 215139100,
87
+ "end": 215140720,
88
+ "sha256": "a6e330a9166ff933b226263ddda384c8a6d03167e3d5bd4d39ce3a1ba1b6a103"
89
+ },
90
+ {
91
+ "id": "managed-settings",
92
+ "start": 205478700,
93
+ "end": 205479600,
94
+ "sha256": "85b72e92961af88f400d9630ace4a42eb61ea55236622c6aa812cf59434aca9e"
95
+ },
96
+ {
97
+ "id": "api-key",
98
+ "start": 215175700,
99
+ "end": 215176620,
100
+ "sha256": "40979b32fca9c0d29f6546378f8105ae1074bd915d4672475822b0209ee839f0"
101
+ },
102
+ {
103
+ "id": "login",
104
+ "start": 203523150,
105
+ "end": 203523800,
106
+ "sha256": "44e632bee5f03f98df47fbfcde92d76dfbbb70cbc59cafa4de797b80c0e83e83"
107
+ },
108
+ {
109
+ "id": "permission-question",
110
+ "start": 208009000,
111
+ "end": 208009700,
112
+ "sha256": "63e00a8b392249ec7c730314f4f95931ff3f8b9d4182f15be6434b5c76b082de"
113
+ },
114
+ {
115
+ "id": "connection-question",
116
+ "start": 209416500,
117
+ "end": 209417050,
118
+ "sha256": "53ac231cb85f5f70d042daca6ee58748d880ed79f98a6e0b15f189aa6080654d"
119
+ },
120
+ {
121
+ "id": "permission-wait",
122
+ "start": 209207500,
123
+ "end": 209208080,
124
+ "sha256": "0f483d8338bdf69daacf8b671bcac4b7bbd95094f99da5d7ef919c36c3e54fa9"
125
+ },
126
+ {
127
+ "id": "confirmation-guide",
128
+ "start": 202713350,
129
+ "end": 202714100,
130
+ "sha256": "830d7821db28f650d2465690ab7d35076768b8c091f3667f945b839af4426555"
131
+ },
132
+ {
133
+ "id": "lower-cancel",
134
+ "start": 197647850,
135
+ "end": 197648260,
136
+ "sha256": "14947f734aa61d45fb13fb17e41294825b70598ca97f782486063995569848fa"
137
+ },
138
+ {
139
+ "id": "selection-dialog",
140
+ "start": 217942250,
141
+ "end": 217943000,
142
+ "sha256": "4a454e26a631e6c39fc490379d94be0e67b973f4fba3421137b9ef82a8e31d34"
143
+ },
144
+ {
145
+ "id": "gateway",
146
+ "start": 203506200,
147
+ "end": 203506740,
148
+ "sha256": "359b0da11cab049a15c9f8411c25a38b5d54d22007fc8439b328893170eb5963"
149
+ },
150
+ {
151
+ "id": "round-border-glyphs",
152
+ "start": 182630499,
153
+ "end": 182630665,
154
+ "sha256": "6b03cfc0d757ebd18a1f171a012e05e78af72dba34cd636f6843272282d14f5d"
155
+ }
156
+ ],
157
+ "note": "Read-only source qualification; no harness execution or copied modules. Marker set is finite, not exhaustive of unknown dialogs.",
158
+ "contract": "post-submitted-completion-v1",
159
+ "layout": "non-Jt (tengu_copper_thistle=false); no runtime flag inspection",
160
+ "acceptedRisk": "Unknown dialogs retaining the composer/footer can be misclassified completed; no subsequent input or readiness inference.",
161
+ "markers": [
162
+ "Accessing workspace:",
163
+ "Quick safety check:",
164
+ "Yes, I trust this folder",
165
+ "No, continue without these permissions",
166
+ "No, exit",
167
+ "WARNING: Loading development channels",
168
+ "I am using this for local development",
169
+ "Managed settings require approval",
170
+ "Yes, I trust these settings",
171
+ "No, exit Claude Code",
172
+ "Do you want to use this API key?",
173
+ "Detected a custom API key in your environment",
174
+ "Select login method:",
175
+ "Do you want to proceed?",
176
+ "Do you want to allow this connection?",
177
+ "Waiting for permission…",
178
+ "Enter to confirm",
179
+ "Esc to cancel",
180
+ "esc to cancel",
181
+ "Enter to select",
182
+ "Esc to exit"
183
+ ],
184
+ "observedRegression": {
185
+ "file": "test/fixtures/launch-prompts/claude-max-started-home-sanitized.txt",
186
+ "classification": "SANITIZED DERIVATIVE, not raw replay or exact spinner matcher",
187
+ "rawBytes": 1217,
188
+ "rawSha256": "bb59a8cf1ef0d1a07c7f298e4c6f51227f12e09e6c8ae16b2eb6bc4248436a93",
189
+ "derivativeBytes": 1156,
190
+ "derivativeSha256": "49c8073ae86da4325dc03587d4ff1568cc39b8f6dda0d451bc776b876adcb6a0",
191
+ "changedRowIndices": [
192
+ 2
193
+ ],
194
+ "sampleHome": "/fixture/home",
195
+ "geometry": {
196
+ "width": 110,
197
+ "height": 35
198
+ },
199
+ "source": "controller-compared-frame",
200
+ "captureTime": null,
201
+ "retainedAt": "2026-10-07T20:10:57.986Z",
202
+ "candidateHead": "57d993fc28681e8f4803d19bf227f19ff43f4aea",
203
+ "meaning": "Observed transient in-progress task turn; not idle, ready, or recurring spinner qualification."
204
+ }
205
+ }
@@ -1,6 +1,7 @@
1
1
  import { constants, openSync, closeSync, readSync, fstatSync, lstatSync, realpathSync, readFileSync } from "node:fs";
2
2
  import { createHash } from "node:crypto";
3
3
  import { homedir } from "node:os";
4
+ import { classifyClaudeLaunchCompletion } from "./launch-prompt-completion.mjs";
4
5
 
5
6
  const ROOT = new URL("./launch-prompt-fixtures/claude-2.1.289-darwin-arm64/", import.meta.url);
6
7
  const MANIFEST = JSON.parse(readFileSync(new URL("manifest.json", ROOT), "utf8"));
@@ -58,7 +59,7 @@ function captured(name) {
58
59
  }
59
60
 
60
61
  // Accepted as a completion-only finite policy, not a general claim about
61
- // Claude's path renderer. Apart from the source-derived billing atom below,
62
+ // Claude's path renderer. Apart from the source-derived header/footer below,
62
63
  // all surrounding bytes remain the captured frame.
63
64
  // Nothing produced here can match a prompt or authorize an input action.
64
65
  function completedFrames(home) {
@@ -79,11 +80,19 @@ function completedFrames(home) {
79
80
  // empty input/status frame remains exact; these are not account captures.
80
81
  // See launch-prompt-fixtures/README.md for the pinned source provenance.
81
82
  const billings = ["API Usage Billing", "Claude Max", "Claude Pro", "Claude Team", "Claude Enterprise"];
82
- return banners.flatMap(name => [...paths].filter(path => prefix.length + path.length <= MANIFEST.geometry.width).flatMap(path => billings.map(billing => {
83
+ return banners.flatMap(name => [...paths].filter(path => prefix.length + path.length <= MANIFEST.geometry.width).flatMap(path => billings.flatMap(billing => {
83
84
  const variant = captured(name).split("\n");
84
85
  variant[1] = "▝▜██████▀ Opus 5.5 · " + billing;
85
86
  variant[2] = prefix + path;
86
- return variant.join("\n");
87
+ const auto = variant.join("\n");
88
+ if (billing !== "Claude Max") return [auto];
89
+ // Source-qualified whole-frame variant, not normalization of pane output:
90
+ // the longer non-default mode label consumes nine columns of the gap.
91
+ const autoFooter = " ⏵⏵ auto mode on (shift+tab to cycle) · ← for agents" + " ".repeat(37) + "◐ medium · /effort";
92
+ const footer = name === banners[0] ? 11 : 10;
93
+ if (variant[footer] !== autoFooter) throw new Error("fixture footer integrity");
94
+ variant[footer] = " ⏵⏵ bypass permissions on (shift+tab to cycle) · ← for agents" + " ".repeat(28) + "◐ medium · /effort";
95
+ return [auto, variant.join("\n")];
87
96
  })));
88
97
  }
89
98
 
@@ -112,9 +121,11 @@ export function qualifyLaunchPromptFixtures({ home, harness, executablePath, arg
112
121
  frame(captured("frame-05-channel-seeded.txt"), "prompt", [], { class: "awebDevelopmentChannel" }),
113
122
  ...[[], ["awebDevelopmentChannel"]].flatMap(after => completedFrames(home).map(text => frame(text, "completed", after))),
114
123
  ] });
124
+ result.classifyCompletion = classifyClaudeLaunchCompletion;
115
125
  }
116
126
  } catch {
117
127
  result.fixtures = [];
128
+ delete result.classifyCompletion;
118
129
  }
119
130
  return result;
120
131
  }
@@ -17,7 +17,7 @@ const wait = ms => Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0,
17
17
  * frames (text, width, height), never a substring or whitespace normalization.
18
18
  * audit receives launch-prompt event data and returns appendEvent's checked result.
19
19
  */
20
- export function createLaunchPromptController({ home, startId, harness, policy = {}, transport, audit,
20
+ export function createLaunchPromptController({ home, startId, harness, policy = {}, transport, audit, retainFrame, classifyCompletion,
21
21
  now = () => performance.now(), sleep = wait, fixtures = [], geometry, validateExecutable, expectedHomeIdentity }) {
22
22
  const selectedHarness = typeof harness === "string" ? { name: harness } : { ...harness };
23
23
  const consent = { ...policy };
@@ -96,6 +96,12 @@ export function createLaunchPromptController({ home, startId, harness, policy =
96
96
  return candidates.length === 1 ? candidates[0] : null;
97
97
  }
98
98
  const exact = (a, b) => sameTarget(a, b) && a.width === b.width && a.height === b.height && a.text === b.text;
99
+ function retain(screen, outcome, recognition) {
100
+ try {
101
+ return retainFrame?.({ home, homeIdentity: { ...identity }, startId, target: { ...target },
102
+ screen: { text: screen.text, width: screen.width, height: screen.height }, outcome, recognition })?.ok === true;
103
+ } catch { return false; } // No raw text, OS diagnostic or private receipt escapes.
104
+ }
99
105
  function observe(t, before) {
100
106
  if (used) return terminal ?? { status: "blocked", answers: [], reason: "blocked: launch authority already used", receipt: [] };
101
107
  used = true;
@@ -140,8 +146,24 @@ export function createLaunchPromptController({ home, startId, harness, policy =
140
146
  sleep(Math.min(INTERVAL, remaining()));
141
147
  continue;
142
148
  }
143
- const match = matching(screen);
144
- if (!match) return finish("blocked", "blocked: unexpected prompt");
149
+ const submitted = answers.length === 1 && answers[0].status === "submitted";
150
+ // The initial exact question contains these same markers: completion
151
+ // policy is reachable ONLY after its sole key was durably submitted.
152
+ const completion = submitted ? classifyCompletion?.(screen) : null;
153
+ const match = completion === "blocked" ? null : matching(screen);
154
+ if (!match && completion === "completed") {
155
+ if (!retain(screen, "completed", "structural")) return finish("incomplete", "private launch evidence retention failed");
156
+ return finish("completed", "recognized launch boundary");
157
+ }
158
+ if (!match) {
159
+ let reason = "blocked: unexpected prompt";
160
+ if (submitted && screen.width === 110 && screen.height === 35 && retainFrame) {
161
+ // This is the exact object just compared, not a later pane capture.
162
+ // Private evidence cannot authorize input or change the launch outcome.
163
+ if (!retain(screen, "blocked", "unmatched")) reason += "; private launch evidence retention failed";
164
+ }
165
+ return finish("blocked", reason);
166
+ }
145
167
  const { fixture, frame } = match;
146
168
  const prior = answers.map(a => a.class);
147
169
  if (frame.kind === "completed") return finish("completed", "recognized launch boundary");
@@ -38,7 +38,7 @@
38
38
  },
39
39
  "oats.framework": {
40
40
  "url": "https://github.com/awebai/oats.git",
41
- "ref": "oats-framework/v1.6.1",
41
+ "ref": "oats-framework/v1.7.0",
42
42
  "path": "oats-package"
43
43
  }
44
44
  },
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@awebai/oats",
3
- "version": "0.46.0",
3
+ "version": "0.47.0",
4
4
  "description": "OATS (Open Agent Team Specification) — durable souls, disposable instances, targetable capability packages, and the runtime-neutral oats CLI/kernel.",
5
5
  "keywords": [
6
6
  "agents",
@@ -60,7 +60,7 @@ members:
60
60
  - git:github.com/acme/agents # the host is a member too
61
61
  - git:github.com/acme/platform
62
62
  packages:
63
- oats.framework: v1.6.1 # bare versions resolve through the official catalog
63
+ oats.framework: v1.7.0 # bare versions resolve through the official catalog
64
64
  oats.okf: v4.1.1
65
65
  defaults:
66
66
  capabilities: { oats.core: { from: package } }
@@ -192,11 +192,22 @@ alone is insufficient. Preview reports policy without terminal input;
192
192
  `--no-launch` never answers. If launch returns `E_SPAWN_INCOMPLETE`, inspect the
193
193
  retained home/pane: do not respawn, repeat keys or treat `launched:false` as an
194
194
  absent process.
195
- Completion additionally accepts source-derived Max/Pro/Team/Enterprise labels
196
- on the exact empty Opus 5.5 / auto / medium frame. This interim support does not
197
- cover task-filled or bypass-permissions screens and does not prove readiness.
195
+ Completion retains the exact empty-frame families and, only after the submitted
196
+ channel Enter, can recognize the qualified bottom composer/effort-footer region.
197
+ See [the bounded completion contract](../../docs/launch-prompt-completion.md).
198
+ Known question markers block first; normal shortcut/interrupt hints do not.
199
+ Unknown dialogs retaining that region may be misclassified completed. Completion
200
+ closes input authority, not a readiness or idle-state promise. Missing effort or
201
+ trailing notices still block. Required private evidence failure is incomplete.
198
202
 
199
203
  Create and spawn only when asked. After the first spawn, the `oats.setup`
200
204
  skills carry the rest of the deployment (messaging, more souls, rebuilds); the
201
205
  `oats.authoring` package teaches writing capabilities, skills and souls. The
202
206
  model in full is `docs/workspaces.md` in the OATS repository.
207
+
208
+ For a retained launch-prompt block after a submitted answer, inspect the existing
209
+ launch event and private frame receipt as described in
210
+ `docs/execution-targets.md#private-unmatched-launch-evidence`. Do not transmit raw
211
+ screen text or treat a saved frame as a new accepted signature. A structurally completed frame is also saved privately. A saved artifact alone
212
+ does not establish completion: verify the final checked launch outcome. Missing
213
+ effort or unsupported footer shapes can block while the pane is active.