[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M79PEWFFK2Z3D0MB2681C6P7":3,"skill-same:01M79PEWFFK2Z3D0MB2681C6P7":710},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":30,"ownerHtmlUrl":31,"ownerBlogUrl":32,"ownerDescription":33,"updatedAt":34,"install":35,"resources":48},"01M79PEWFFK2Z3D0MB2681C6P7","boucle",null,"The boucle protocol — the labels, markers, state machine and handoff rules a\nboucle loop speaks. Use when interacting with a boucle loop on any consumer\nrepository: reading or writing boucle:* labels, emitting or parsing\n\u003C!-- boucle:* --> markers, posting a SHA-anchored verdict, driving an issue\nthrough triage\u002Fworker\u002Freviewer\u002Fe2e, or answering \"what is this loop waiting\non\". Runtime-agnostic and forge-agnostic (GitLab and GitHub).","ankaboot-source\u002Fboucle","ankaboot-source","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F88089122?v=4","01M79P4HZGM4NXFWV0SB0HGQ51","ankaboot","01M79P4J0GNNZ7E7F9B0RZNSNX",2,"AGPL-3.0","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: boucle\ndescription: |\n  The boucle protocol — the labels, markers, state machine and handoff rules a\n  boucle loop speaks. Use when interacting with a boucle loop on any consumer\n  repository: reading or writing boucle:* labels, emitting or parsing\n  \u003C!-- boucle:* --> markers, posting a SHA-anchored verdict, driving an issue\n  through triage\u002Fworker\u002Freviewer\u002Fe2e, or answering \"what is this loop waiting\n  on\". Runtime-agnostic and forge-agnostic (GitLab and GitHub).\n---\n\n# SKILL.md — Boucle protocol\n\n> **Source of truth for the boucle protocol.** This file is the behavioral\n> contract: the labels, markers, state machine, and handoff rules that boucle\n> speaks. It is runtime-agnostic (does not reference jcode internals) and\n> forge-agnostic (covers GitLab + GitHub). A local harness (opencode, jcode,\n> pi, or any agent runtime that reads markdown) loads this skill to \"speak\n> boucle\" — interact with a boucle loop on any consumer repo correctly.\n>\n> **Projection at install:** `bin\u002Fsetup` symlinks `.jcode\u002Fskills\u002Fboucle\u002FSKILL.md`\n> → `..\u002F..\u002FSKILL.md` so the consumer's harness loads it from the conventional\n> skills directory. The source is this file; the symlink is the projection.\n>\n> **Enforcement:** `bin\u002Fcheck-doc-sync` validates that the labels and markers\n> in the engine code match this file. A label or marker in code that is absent\n> here fails CI red — this file is the spec, not a description.\n>\n> **Relationship to other charters:**\n> - `AGENTS.md` — contribution conventions + mandatory principles. `LESSONS.yml` — lessons (incident\n>   catalog). Lessons that state a current protocol invariant cross-reference\n>   this file (`See SKILL.md §\u003Cid>`); the normative text lives here.\n> - `ARCHITECTURE.md` — engine implementation (code structure, forge adapters,\n>   CI stages). The *how*, not the *what*.\n> - `CONTEXT.md` — identity, audience, philosophy, constraints. The *why*.\n> - `README.md` — overview, getting started. The *entry*.\n\n---\n\n## 1. Invariants\n\nThe protocol rests on ten invariants. Every agent (CI or local harness) and\nevery human interacting with boucle MUST honor them. The LESSONS.yml lessons are\nincident catalogs that instantiate these invariants; the normative statement\nlives here.\n\n### I1 — Forge-native\n\nBoucle lives in the forge. NEVER introduce a new frontend, a server, or a\ncomputer to keep running. The forge is the UI; labels are the state; comments\nare the channel. (CONTEXT.md §7, LESSONS.yml lesson #55.)\n\n### I2 — Label-driven state machine\n\nLabels are the source of truth. No external database, no separate state API.\nThe state machine is driven by `boucle:*` labels (detail axis) + `boucle::status::*`\nlabels (gross axis: `bot`\u002F`human`\u002F`done`). A label change IS the state\ntransition. (CONTEXT.md §7.)\n\n### I3 — Async by design\n\nThe human is not in front of the screen. Work happens asynchronously, driven\nby labels and comments. The loop does not wait for a human reply — it progresses\nto the next gate and parks at a human-readable state (`boucle:spec-review`,\n`boucle:approval`, `boucle:human`). (CONTEXT.md §1.)\n\n### I4 — Post-early\n\nPost the comment or verdict FIRST, then refine. An incomplete draft posted is\nALWAYS better than a refinement never posted. Step-budget waste (the agent\nexhausts its budget without posting) is bug #1. (LESSONS.yml lesson #1, #2, #5.)\n\n**\"Post early\" means minimal but meaningful, not an empty placeholder.** A\ndraft MUST contain at least the content the human needs to act on it — for\ntriage, a rough `## Analysis` section (2-3 sentences restating the issue) and\na `## Metadata` section carrying a `Disposition`. An empty placeholder (\"DRAFT — first-pass triage,\nrefining next.\") is noise the human cannot act on: no analysis, no questions,\nno criteria. The agent MUST read the issue body (already in its prompt as\n`$BOUCLE_ISSUE_BODY`) before posting — \"post early\" does NOT mean \"post\nbefore reading the issue\". (LESSONS.yml lesson #99.)\n\n### I5 — Idempotence\n\nAll `bin\u002F*` scripts and label writes MUST be idempotent. Re-running a script\nproduces no additional side effects. A label PUT that does not change the label\nset is skipped (the forge records a Resource Label Event on every PUT, even a\nno-op — CONTEXT.md §8). (LESSONS.yml lesson #4.)\n\n### I6 — SHA-anchored verdicts\n\nReviewer and e2e verdicts MUST include the commit SHA as bare hex: no quotes, no\nwhitespace, no angle brackets. Exact format:\n`\u003C!-- boucle:verdict v=1 role=reviewer sha=abc123def456 -->`. The CI parser\nFAILS if the format is not respected. (LESSONS.yml lesson #6, #41, #47.)\n\n### I7 — Marker-based self-recognition\n\nBoucle recognizes its own writes by an invisible stamp\n(`\u003C!-- boucle:agent -->`), NEVER by the actor's identity. A comment posted\nwithout the stamp is treated as a human reply and routed. This is mono-user-safe\n(when bot and human share an account, the marker discriminates; the actor does\nnot). Every comment a harness posts MUST carry the stamp, or dispatch will treat\nit as a human reply and re-route. (LESSONS.yml lesson #55.)\n\nThe same marker discriminates note authorship in prompts: CI MUST tag every\nnote injected into an agent prompt `[\u003Cauthor> — boucle]` when its body carries\nthe stamp, `[\u003Cauthor> — human]` otherwise, and agents MUST classify authorship\nby that tag — NEVER by the posting account (in mono-user mode the account is\nshared, so identity discriminates nothing; a human amendment posted under the\nbot's account is still a binding amendment, lesson #111). (LESSONS.yml lesson\n#111.)\n\n### I8 — Doc-as-code\n\nA doc that describes a system that no longer exists is a bug. Charter docs are\nmaintained as part of each work cycle: triage identifies impacted docs, worker\nupdates them in the same MR, reviewer verifies conformance, e2e verifies\nproduction match. `bin\u002Fcheck-doc-sync` enforces code ↔ SKILL.md sync in CI.\n(AGENTS.md \"Documentation self-maintenance\".)\n\n### I9 — Upstream-first\n\nFix upstream (in boucle) FIRST, then update the consumer, then remediate\nexisting data. NEVER patch a consumer to work around a boucle defect. NEVER\nintroduce a local workaround that won't be reported upstream. (CONTEXT.md §7,\n`.jcode\u002FUPSTREAM-FIX-WORKFLOW.md`.)\n\n### I10 — Serial merge\n\nMerges are serialized via `resource_group: boucle-merge`. Each rebase is\nagainst a `master`\u002F`main` that includes previously-merged MRs. NEVER\nparallelize merges — a concurrent rebase against a stale branch produces\nconflicts and race conditions. (CONTEXT.md §7, LESSONS.yml lesson #8.)\n\n---\n\n## 2. State machine\n\nThe state machine is two-axis: a **detail** label (`boucle:\u003Cstate>`) that\nanswers \"what is the loop doing?\" and a **gross** label\n(`boucle::status::\u003Cowner>`) that answers \"whose side is this on?\". In\nmono-user mode (`BOUCLE_MONO_USER=true`) the gross axis is dropped (one actor\nowns both sides — the question is meaningless).\n\n### 2.1 Detail labels\n\n| Label | Meaning | Owner (gross) |\n|---|---|---|\n| `boucle:triage` | Awaiting triage analysis | bot |\n| `boucle:needs-info` | Triage needs more info from the reporter | human |\n| `boucle:spec-review` | Spec validated by triage, awaiting human spec approval | human |\n| `boucle:approved` | Human approved the spec (GitHub: label added by author) | bot |\n| `boucle:todo` | Spec approved, queued for the worker | bot |\n| `boucle:working` | Worker is running | bot |\n| `boucle:review` | Worker shipped, awaiting reviewer verdict | bot |\n| `boucle:approval` | Reviewer PASSed, MR ready, awaiting human MR approval | human |\n| `boucle:merging` | Merger is running (rebase + merge) | bot |\n| `boucle:done` | Loop complete (merged + e2e PASS) | done |\n| `boucle:human` | Escalated to a human (iteration cap, unclear criteria, destructive change, etc.) | human |\n| `boucle:blocked` | Waiting on a dependency (sub-issue not closed) | bot |\n| `boucle:split` | Parent issue split into sub-issues, waiting for them | bot |\n| `boucle:dnd` | Transient flag: spec gate auto-validated during DND window | (rides along) |\n| `boucle:autonomous` | Transient flag: spec gate skipped per-issue opt-in | (rides along) |\n| `boucle:recurring` | Context tag: issue is part of a recurring bug class (non-blocking, survives transitions) | (rides along) |\n| `boucle:board` | The status-board issue (never dispatched) | — |\n| `boucle:scheduled` | Issue created by a schedule (cron template) | — |\n| `boucle:retro` | The retro's local per-run report issue (never dispatched) | — |\n| `boucle:report` | Upstream retro proposal — anonymized case or plan, created on `BOUCLE_UPSTREAM_REPO` (never dispatched) | — |\n\n**Removed (dead labels, pruned from `bin\u002Fsetup`):** `boucle:spec-approved`.\nMR approval uses the native forge Approve button (GitLab) or the\n`boucle:approved` label (GitHub — no native Approve button on issues).\nSpec approval uses a canonical emoji reaction (👍 ❤️ 🎉 🚀) on\n`boucle:spec-review`, the `boucle:approved` label, or the magic word\n`approved` — a text reply amends the spec and re-triggers triage, it\ndoes NOT approve.\n\n### 2.2 Gross labels\n\n| Label | Meaning |\n|---|---|\n| `boucle::status::bot` | The loop owns the next action (issue assigned to bot) |\n| `boucle::status::human` | A human owns the next action (issue assigned to human reporter) |\n| `boucle::status::done` | Terminal (loop complete) |\n\n### 2.3 State diagram\n\n```mermaid\nstateDiagram-v2\n    [*] --> triage: boucle:triage added \u002F bot assigned\n    triage --> needs_info: triage needs more info\n    triage --> spec_review: triage validated spec (Size S)\n    triage --> todo: triage validated spec (auto \u002F DND \u002F autonomous)\n    triage --> human: Size L \u002F unclear criteria \u002F destructive\n    triage --> split: issue too big, split into sub-issues\n    needs_info --> triage: human replied (note on needs-info)\n    spec_review --> todo: human approved spec (👍 ❤️ 🎉 🚀 emoji)\n    spec_review --> triage: human replied with amendment (re-triage)\n    spec_review --> human: human rejected \u002F no response\n    todo --> working: worker started\n    working --> review: worker shipped code\n    working --> todo: worker no-changes \u002F build-fail (retry, iter \u003C max)\n    working --> human: worker iteration cap \u002F API down (exit 4)\n    review --> approval: reviewer PASS\n    review --> todo: reviewer FAIL (retry, iter \u003C max)\n    review --> human: reviewer FAIL (iter cap) \u002F UNCERTAIN \u002F no verdict\n    approval --> merging: human approved MR (native Approve button)\n    approval --> review: MR updated (push to boucle\u002F\u003Ciid>-\u003Cslug>)\n    approval --> human: MR closed without merge\n    merging --> done: merge succeeded + e2e PASS\n    merging --> human: merge conflict \u002F not mergeable\n    done --> [*]\n    human --> triage: human re-assigns bot (BOT_JUST_ASSIGNED)\n    human --> [*]: human closes issue\n    split --> triage: all sub-issues closed (parent re-queued)\n    blocked --> todo: dependency closed (unblock)\n```\n\n**Entry is opt-in.** An issue opened with no `boucle:` label does NOT enter\nthe state machine: it has no boucle state at all, and the loop never touches\nit. A human hands it over by adding `boucle:triage` or by assigning the bot\n(`BOT_JUST_ASSIGNED`) — the two edges out of `[*]` above. Issues boucle opens\nfor itself (schedules, triage sub-issues, the e2e follow-up) pass\n`boucle:triage` at creation, so they enter through the same edge as anything\na human hands over. Setting `BOUCLE_ENTRY_MODE=auto` restores the legacy\nbehaviour, where the `issue opened` webhook alone was enough — appropriate\nonly for a tracker that exists for boucle and nothing else.\n\n### 2.4 Transition table\n\nThe transition table is derived from the handoff primitives in §4. Every\ntransition is effected by `set_boucle_label \u003Ciid> \u003Cdetail> \u003Cgross>` (lib\u002Fboucle.sh:626),\nwhich preserves non-boucle labels, writes the detail+gross pair idempotently,\nreassigns the issue (bot on `boucle::status::bot`, human reporter on\n`boucle::status::human`), and fires the outbound notification on the\ntransition (never on the state).\n\n---\n\n## 3. Marker reference\n\nBoucle communicates via invisible HTML-comment markers stamped on issue\u002FMR\ncomments and via structural sections in comment bodies. The markers are\nmachine-readable; the structural sections are detected by pattern. A local\nharness MUST use the markers correctly or dispatch will misroute its writes.\n\n### 3.1 Self-recognition marker\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:agent -->` | bare HTML comment | `stamp_agent_marker` (bin\u002Fforge\u002Fcommon.sh:81) on every `forge_issue_note` \u002F `forge_mr_note` | `has_agent_marker` (bin\u002Fforge\u002Fcommon.sh:92) in dispatch.sh:99-103 | Distinguishes boucle's own writes from human replies. The primary anti-loop guard (invariant I7). A harness posting a comment MUST carry this stamp or dispatch treats it as a human reply and re-routes. |\n\n### 3.2 Triage markers\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:triage v=1 -->` | `v=1` (no attrs) | triage agent (final comment); draft promoted by triage.sh:178 (`sed s\u002Fdraft role=triage\u002Ftriage v=1\u002F`) | triage.sh:163 (awk), collapse-duplicate-notes:71-72 (jq structural filter) | Marks a final triage comment. CI parser acts on it (sets disposition label, assigns, pauses). |\n| `\u003C!-- boucle:draft role=triage -->` | `role=triage` | triage agent (first-pass draft) | collapse-duplicate-notes:73 (filter), triage.sh:178 (promotion source) | Marks a draft triage comment. CI parser does NOT act on drafts; log-scraping fallback promotes to final when the agent exhausts its steps. |\n| `\u003C!-- boucle:obligations v=1 -->` | `v=1` | triage.sh:213 | (informational — marks the obligations section of a triage comment) | Marks the obligations block in a triage comment (what the human must do next). |\n| `\u003C!-- boucle:needs-info v=1 reason=no-key -->` | `v=1 reason=no-key` | triage.sh:92 (no-LLM-key gate) | triage.sh:55 (jq filter, idempotency check) | Marks a needs-info comment posted when the LLM API key is missing. Used for idempotency: triage checks if a no-key needs-info was already posted before re-posting. |\n| `\u003C!-- boucle:needs-info v=1 reason=profile -->` | `v=1 reason=profile` | lib\u002Fboucle-ci\u002Ftriage.sh (first-issue profile determination, #186) | lib\u002Fboucle-ci\u002Ftriage.sh (jq filter, idempotency check) | Marks the first-issue project-profile determination comment (detected `infra` on an auto\u002Funset install). The human confirms by 👍\u002F👎 reaction or a `profile: site\\|infra` reply; no agent run is burned while it waits. |\n| `\u003C!-- boucle:profile-applied v=1 profile=infra -->` | `v=1 profile=infra\\|site` | lib\u002Fboucle-ci\u002Ftriage.sh (profile apply path, #186) | lib\u002Fboucle-ci\u002Ftriage.sh (jq filter, idempotency check) | Marks the first-issue profile determination as applied (infra preset set, or site recorded). Once present, triage never re-posts a determination for the issue. |\n| `\u003C!-- boucle:diagram-missing v=1 kinds=\u003Ckinds> -->` | `v=1 kinds=\u003Cclosed-set>` | gates.sh:169 (spec-completeness gate) | gates.sh (re-trigger triage) | Marks a spec that declares structural impacts (Size M\u002FL) but is missing the required `## Diagram` section. CI re-triggers triage to add only the missing diagram. |\n| `\u003C!-- boucle:preview-missing v=1 kinds=\u003Ckinds> -->` | `v=1 kinds=\u003Cclosed-set>` | gates.sh:227 (spec-completeness gate) | gates.sh (re-trigger triage) | Marks a spec that declares visual impacts (Size M\u002FL) but is missing the required `preview.html` \u002F `RENDER_REQUEST`. CI re-triggers triage to add only the missing preview. |\n| `\u003C!-- boucle:diagram-invalid v=1 rounds=\u003Cn> -->` | `v=1 rounds=\u003Cn>` or `v=1 state=exhausted rounds=\u003Cn>` | gates.sh (check_diagram_syntax_gate) | gates.sh (retry-bound count) | Marks a spec whose Mermaid diagram the real parser rejects — the forge would render it as an error box. CI re-triggers triage to fix ONLY the diagram. After `BOUCLE_DIAGRAM_SYNTAX_MAX_RETRIES` rounds the marker carries `state=exhausted` and the spec goes to the human as-is instead of looping. |\n\n### 3.3 Verdict markers (reviewer \u002F e2e)\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:verdict v=1 role=reviewer sha=\u003Chex> -->` | `v=1 role=reviewer sha=\u003Cshort-sha>` | reviewer agent (final verdict); draft promoted by reviewer.sh:381 | reviewer.sh:357 (SHA-anchored), :362 (SHA-unanchored fallback) | Marks a final reviewer verdict. SHA MUST be bare hex (no quotes\u002Fwhitespace\u002Fangle brackets) — invariant I6. |\n| `\u003C!-- boucle:verdict v=1 role=e2e sha=\u003Chex> -->` | `v=1 role=e2e sha=\u003Cshort-sha>` | e2e agent (final verdict); draft promoted by e2e.sh:192 | e2e.sh:179 | Marks a final e2e verdict. Same SHA contract. |\n| `\u003C!-- boucle:draft role=reviewer -->` | `role=reviewer` | reviewer agent (first-pass draft) | collapse-duplicate-notes:79, reviewer.sh:381 (promotion source) | Draft reviewer verdict. Parser does NOT act; log-scraping promotes on step exhaustion. |\n| `\u003C!-- boucle:draft role=e2e -->` | `role=e2e` | e2e agent (first-pass draft) | collapse-duplicate-notes:86, e2e.sh:187 (promotion source) | Draft e2e verdict. Same as above. |\n\n### 3.4 Dependency \u002F hierarchy markers\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:depends-on iids=N,M -->` | `iids=\u003Ccomma-separated-IIDs>` | triage.sh:755 (on each dependent sub-issue) | bin\u002Flib\u002Fdepends-on.sh:38 (grep), dispatch.sh:427 (gate check) | Declares sub-issue dependencies. Dispatch gates the worker until all dep IIDs are closed. |\n| `\u003C!-- boucle:split-parent iids=N,M -->` | `iids=\u003Ccomma-separated-IIDs>` | triage (split operation) | catchup.sh:66, doctor.sh:746, e2e.sh:75, triage.sh:518,520 (idempotency guard) | Marks the parent issue of a split with its sub-issue IIDs. Used to cascade closure and detect duplicate splits. |\n| `\u003C!-- boucle:blocked v=1 iids=N,M -->` | `v=1 iids=\u003Ccomma-separated-IIDs>` | dispatch.sh:461, boucle.sh:1136 | (informational — marks the blocked note) | Posted on a sub-issue blocked by open siblings. The label `boucle:blocked` is the state; this marker is the note's machine-readable header. |\n| `\u003C!-- boucle:sibling-blocked v=1 sib=N -->` | `v=1 sib=\u003Cactive-sibling-IID>` | dispatch.sh:508, .gitlab-ci.yml:997 | catchup.sh:86,94, e2e.sh:95,102 (jq parse) | Posted when a sibling sub-issue is still active (working\u002Freview\u002Fapproval). The issue starts automatically once the sibling reaches done\u002Fclosed — sibling issues share the same domain and run one at a time to avoid merge conflicts. |\n| `\u003C!-- boucle:unblocked v=1 by=N -->` | `v=1 by=\u003Cclosed-dep-IID>` | catchup.sh:100, e2e.sh:109 | (informational — marks the unblock note) | Posted when a dependency closes and the worker starts. |\n| `\u003C!-- boucle:e2e-origin v=1 iid=N -->` | `v=1 iid=\u003Corigin-IID>` | e2e.sh:267 (on follow-up issues) | boucle.sh:705 (parse, to resolve the original issue from a follow-up) | Carries the original issue IID on a follow-up issue created by e2e FAIL, so the loop can cascade back. |\n| `\u003C!-- boucle:e2e-fail v=1 iid=N followup=M -->` | `v=1 iid=\u003Corigin-IID> followup=\u003Cfollowup-IID>` | e2e.sh:291 (on follow-up issue created by e2e FAIL) | (informational — links follow-up to origin) | Posted on the follow-up issue created when e2e FAILs, linking it to the original issue for cascade. |\n| `\u003C!-- boucle:e2e-escalation v=1 iid=N verdict=X -->` | `v=1 iid=\u003CIID> verdict=\u003Cuncertain\\|empty>` | e2e.sh:321,357 (on e2e UNCERTAIN or no-verdict) | (informational — structured escalation) | Posted when e2e escalates to human due to UNCERTAIN verdict or no verdict (step exhaustion). Records the verdict reason. |\n| `\u003C!-- boucle:sub-issue v=1 -->` | `v=1` | triage (split operation) | triage.sh:566 (skip in parent-body parsing) | Marks a sub-issue body. Used to avoid parsing sub-issue content as parent-issue content. |\n| `\u003C!-- boucle:recurring v=1 refs=N,M -->` | `v=1 refs=\u003Ccomma-separated-IIDs>` | triage agent (optional `## Recurring theme` section) | triage.sh (parse + label), worker.sh (inject refs as context) | Recurring-theme declaration: flags the issue as part of a recurring class and cites prior closed issues. CI applies `boucle:recurring` (context tag surviving transitions). Worker receives prior-issue summaries to find the root cause. Non-blocking: never gates. |\n| `\u003C!-- boucle:impacts v=1 kinds=... -->` | `v=1 kinds=\u003Ccomma-separated-kinds>` | triage agent (`## Metadata` section in the spec comment) | triage.sh (parse_impacts_marker), gates.sh (check_diagram_gate, check_preview_gate) | Structural\u002Fvisual impact declaration (LESSONS.yml lesson #104): the triage declares which impacts this issue has from a closed set (architecture, data-model, process, state-machine, data-flow, deployment, ui, ux, design, none). CI cross-checks: a structural kind without a `\u003C!-- boucle:diagram v=1 -->` marker → gate fails → triage re-triggered; a visual kind (ui\u002Fux\u002Fdesign) without a `preview.html` + `RENDER_REQUEST` → gate fails → triage re-triggered. Drives the deterministic diagram\u002Fpreview gate. |\n| `\u003C!-- boucle:diagram v=1 types=... -->` | `v=1 types=\u003Ccomma-separated-Mermaid-block-types>` | triage agent (`## Diagram` section in the spec comment) | triage.sh (parse_diagram_marker), gates.sh (check_diagram_gate) | Diagram declaration: lists the Mermaid block types used in the `## Diagram` section (e.g. `erDiagram,flowchart`). CI cross-checks this marker against the `Impacts` field of `## Metadata`: a structural impact declared there without a `\u003C!-- boucle:diagram v=1 -->` marker → gate fails → triage re-triggered. The Mermaid theme block (boucle.dev light\u002Ftransparent) is defined in `templates\u002Fdiagram-theme.md` — the static source of truth. |\n| `\u003C!-- boucle:diagram-missing v=1 kinds=... -->` | `v=1 kinds=\u003Ccomma-separated-kinds>` | gates.sh:169 (check_diagram_gate) | (informational — re-trigger note) | Posted when a spec declares structural impacts (`## Metadata`) but lacks the required `## Diagram` section. Triage is re-triggered to add ONLY the missing diagram — it must re-read its previous spec in the Prior discussion, not re-analyze from scratch. |\n| `\u003C!-- boucle:preview-missing v=1 kinds=... -->` | `v=1 kinds=\u003Ccomma-separated-kinds>` | gates.sh:227 (check_preview_gate) | (informational — re-trigger note) | Posted when a spec declares visual impacts (`## Metadata`) but lacks the required `preview.html` \u002F `RENDER_REQUEST`. Triage is re-triggered to add ONLY the missing preview — it must re-read its previous spec in the Prior discussion, not re-analyze from scratch. |\n| `\u003C!-- boucle:diagram-invalid v=1 rounds=... -->` | `v=1 rounds=\u003Cn>` (+ `state=exhausted` once the bound is spent) | gates.sh (check_diagram_syntax_gate) | gates.sh (counts its own markers for the retry bound) | Diagram SYNTAX verdict (LESSONS.yml lesson #109). `bin\u002Fcheck-mermaid` runs every ```mermaid fence in the spec through the real Mermaid parser; a parse error means the forge renders an error box instead of a diagram, so the gate blocks spec-review and re-triggers triage with the parser's own message. Fail-open when the parser is unavailable (no node, no modules, no network) — infrastructure never blocks the loop. |\n\n### 3.5 Operational markers\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:board v=1 -->` | `v=1` | boucle.sh:268 (status board body) | (informational — identifies the board issue) | Marks the status-board issue body. Dispatch skips `boucle:board`-labelled issues; this marker is the body's stamp. |\n| `\u003C!-- boucle:catchup v=1 iid=N state=X target=Y -->` | `v=1 iid=\u003CIID> state=\u003Cfrom> target=\u003Cto>` | catchup.sh (audit note on direct merge) | (informational — audit trail) | Posted when an MR is merged directly (not via the approval flow). Records the state at merge time for audit. The `target` field is the terminal state the loop WOULD have applied pre-e2e-on-direct-merge; the actual terminal state is decided by the e2e verdict (PASS→done, FAIL\u002FUNCERTAIN→human). |\n| `\u003C!-- boucle:commit sha=\u003Chex> -->` | `sha=\u003Cshort-sha>` | worker.sh:421-422 (in MR description + build marker) | (preview freshness assertion — worker.sh) | Anchors the MR\u002Fbuild to a commit SHA for preview freshness check (invariant I6-adjacent). |\n| `\u003C!-- boucle:diagnostic v=1 iid=N class=X trigger=Y -->` | `v=1 iid=\u003CIID> class=\u003Cfailure-class> trigger=\u003Ctrigger>` | boucle.sh:475 (escalation diagnostic) | (informational — structured escalation) | Replaces the generic \"human intervention needed\" note with a structured diagnostic (failure class + evidence + recommended action). |\n| `\u003C!-- boucle:schedule id=\u003Cname> -->` | `id=\u003Cschedule-name>` | boucle.sh:201 (on scheduled issues) | boucle.sh (dedup — last firing marker) | Marks a scheduled issue with its template name. Used to prevent duplicate firings across sweeps. |\n| `\u003C!-- boucle:state v=1 -->` | `v=1` | boucle.sh:436 (`boucle_state_save`, on worker exit) | boucle.sh:462 (`boucle_state_restore`, on a cold state cache) | Per-issue loop state (iteration log, approach, tried-and-rejected, `last-outcome`) persisted on the issue so it survives an ephemeral runner, where `BOUCLE_STATE_CACHE` does not. Collapsed, and excluded from the notes injected into agent prompts (bin\u002Fjc) so it is never re-billed as prompt input. |\n| `\u003C!-- boucle:conflict-retry N -->` | `N=\u003Cretry-count>` | boucle.sh:1099 (on re-trigger after rebase conflict) | boucle.sh:1085 (jq parse, count retries) | Posted on each re-trigger after a rebase conflict. Bounded retry count — after N retries, the conflict is handed to the worker agent for resolution. |\n| `\u003C!-- boucle:file-blocked v=1 on=N paths=... -->` | `v=1 on=\u003Cactive-issue> paths=\u003Coverlap>` | gates.sh:106 (check_file_gate — defer on file overlap) | gates.sh:192 (unblock path, last marker) | The file-impact gate (LESSONS.yml lesson #62): posted when a worker would edit files claimed by an in-flight sibling issue. The unblock path fires when the named blocker closes. |\n| `\u003C!-- boucle:files v=1 paths=... -->` | `v=1 paths=\u003Cimpacted-files>` | triage agent (embedded in the spec comment's `## Metadata` section); worker job (lib\u002Fboucle-ci\u002Fworker.sh, refresh in a separate note after the branch diff) | worker job (jq, find last marker note excluding the spec comment) | File-impact declaration (LESSONS.yml lesson #62): the triage predicts the impacted files inside its spec comment, the worker refreshes the claim in a separate machine note with the actual branch diff. Consumed by `check_file_gate` to defer parallel workers on overlap. |\n| `\u003C!-- boucle:evidence-pack v=1 -->` | `v=1` | bin\u002Fbuild-evidence-pack (header first line of the evidence pack) | — | Marks the auto-generated evidence pack (charter doc snapshot + diff brief) attached to escalations. |\n| `\u003C!-- boucle:allow-list v=1 user=\u003Cusername> -->` | `v=1 user=\u003Cauthor-username>` | boucle.sh (check_allow_list_gate — rejection note) | — | Posted on an issue whose resolved human reporter is not in `BOUCLE_ALLOWED_USERS`. The issue is not accepted by the loop (no role triggered). Fail-open when the variable is unset (legacy). |\n| `\u003C!-- boucle:interactive v=1 -->` | `v=1` | bin\u002Fboucle (cmd_pause, cmd_resume, cmd_restart — interactive harness) | — | Marks a comment posted by the interactive harness (`boucle pause`\u002F`resume`\u002F`restart`). Distinguishes a voluntary human takeover from a `boucle:human` escalation. Informational — dispatch does not act on it. |\n| `\u003C!-- boucle:approval-request v=1 -->` | `v=1` (no attrs) | reviewer agent (posted on the PR via `forge_mr_note` when `boucle_mono_user` is true) | doctor (jq: `select(.body \\| contains(\"boucle:approval-request\"))`, then `forge_note_reactions` polls for a canonical approval emoji) | Mono-user PR approval gate: GitHub blocks self-review, so the reviewer posts this note on the PR and the human reacts 👍 to approve. The doctor's `doctor_mr_approval_emoji` polls the note's reactions and triggers the merger when a non-bot user reacted with a canonical approval emoji (👍 ❤️ 🚀 🎉). Replaces the old `VERDICT: PASS` auto-merge (no human signal). |\n| `\u003C!-- boucle:opt-out v=1 -->` | `v=1` (no attrs) | dispatch.sh:929 (human label-removal webhook, #124) | dispatch.sh:926 (idempotency — marker already present), doctor.sh:499 (unlabeled scan skip) | Posted when a human removes the `boucle:*` labels (GitLab `.changes.labels.previous` \u002F GitHub `unlabeled` event): the explicit \"leave the loop\" signal. Idempotent — later events no-op against it. The doctor's unlabeled scan skips tombstones so the issue is never re-appropriated; re-adding `boucle:triage` or assigning the bot re-enters the loop normally. |\n\n### 3.6 Markers NOT in code (do not emit)\n\n| Marker | Status | Note |\n|---|---|---|\n| *(none currently)* | — | Keep this table empty — a marker that is documented here but emitted by code FAILs `bin\u002Fcheck-doc-sync`. |\n\n### 3.7 Structural signals (not HTML comments)\n\n| Signal | Format | Parsed by | Purpose |\n|---|---|---|---|\n| `## TL;DR` + `## Metadata` | two section headers in the same comment | collapse-duplicate-notes:71-72, triage.sh (PRE_RUN_TRIAGE_ID \u002F COMMENT filters) | Structural final-comment detection for triage. A draft only has `## Metadata`; a final has both `## TL;DR` and `## Metadata`. Defense-in-depth: even if the agent uses the wrong marker, the parser won't act on a draft that lacks `## TL;DR`. Specs posted before the section merge carry a standalone `## Disposition` instead of `## Metadata` — the filters accept either. |\n| `## Parent issue\\n#N` | section header + issue reference | boucle.sh:699,768,774 (awk) | Resolves the parent IID from a sub-issue body. Used by `resolve_reporter_id`, `fetch-issue-attachments`, `maybe_close_parent`. |\n| `## Depends on` | section header | bin\u002Flib\u002Fdepends-on.sh:57,61 (awk fallback) | Fallback parsing of dependencies when the `\u003C!-- boucle:depends-on -->` marker is absent. Mirrors the `## Parent issue` pattern. |\n| `## Approach` | section header in `state.md` | worker.sh:482 (sed extraction), :551 (MR description) | The worker's implementation approach, extracted from `state.md` into the MR description. The reviewer reads it to verify doc conformance. |\n| `VERDICT: PASS\\|FAIL\\|UNCERTAIN` | line-anchored (`^VERDICT:`) | reviewer.sh:299,320,358,363; e2e.sh (same pattern) | The verdict line. MUST be start-of-line anchored in greps (LESSONS.yml lesson #41) — an unanchored grep matches shell traces containing the substring. |\n| `BOT_JUST_ASSIGNED` | assignee-change detection | dispatch.sh:504-539 | The real \"re-queue after boucle:human\" mechanism (§4.1). Detected from `.changes.assignees` on the `issue update` webhook, not from a comment. |\n| Emoji `thumbsup` `heart` `rocket` `tada` | emoji award on a Note | dispatch.sh (`BOUCLE_SPEC_APPROVAL_EMOJIS=\"thumbsup heart rocket tada\"`) | Spec approval + needs-info re-trigger. Valid on an issue at `boucle:spec-review` (approve spec) or `boucle:needs-info` (re-trigger after action). MR approval is the native Approve button, NOT an emoji. |\n| Emoji `eyes` (👀) | emoji award on an **Issue** | written by triage.sh (`ack_issue_taken`, `BOUCLE_ACK_EMOJI=\"eyes\"`); discarded by dispatch.sh | **Bot → human** acknowledgement: triage has picked the issue up. The only signal boucle writes as a reaction; deliberately outside the spec-approval set so it can never read as approval. Carries no routing meaning in either direction — dispatch skips issue-level 👀 events (§4.8). |\n\n---\n\n## 4. Handoff protocol\n\nThe handoff protocol defines how a human (via the forge UI or a local harness)\ninteracts with the loop. Every handoff is a webhook event routed by dispatch.\nThe primitives below are extracted from `lib\u002Fboucle-ci\u002Fdispatch.sh` — they are\nthe real mechanisms, not invented verbs.\n\n### 4.1 Re-queue after `boucle:human` — bot reassignment\n\n**`boucle:human` is a dead-end for comments.** A plain note on an issue at\n`boucle:human` matches no dispatch branch and falls through to `exit 0` —\nnothing happens. The ONLY re-queue path is **re-assigning the issue to the bot**.\n\nWhen a human assigns the bot to an issue (any idle state: `boucle:human`,\n`boucle:needs-info` without reply, unlabeled, `boucle:todo`, `boucle:spec-review`),\nthe `issue update` webhook fires with an assignee change. Dispatch detects\n`BOT_JUST_ASSIGNED` (dispatch.sh:504-539) and routes to triage from any idle\nstate. This is the explicit \"reopen and resume\" signal.\n\n**A local harness that wants to re-queue a `boucle:human` issue MUST\nre-assign the bot, not post a comment.** (LESSONS.yml lesson #11, #44.)\n\n### 4.2 Spec approval — emoji only; a reply amends\n\nTo approve a spec (issue at `boucle:spec-review`), the human awards a\n**`thumbsup` `heart` `rocket` or `tada` emoji** on the triage\u002Fspec-review\nnote (dispatch.sh, gated by `BOUCLE_SPEC_APPROVAL_EMOJIS=\"thumbsup heart\nrocket tada\"`). The emoji triggers `chain_to_role worker`. An emoji by the\nbot account is skipped — only human emoji count.\n\nA **text reply is NOT an approval** — it is a correction\u002Famendment. A reply\nre-triggers **triage** (`chain_to_role triage`), which re-reads the human's\nnotes and posts an **updated spec** back at `boucle:spec-review` for another\napproval round. The triage validation message says so explicitly: *“A reply\nnever approves the spec.”* A bot-authored note (carrying\n`\u003C!-- boucle:agent -->`) is skipped before routing.\n\nThe doctor's orphaned-`boucle:spec-review` recovery mirrors this contract:\nan emoji re-triggers the worker; a reply re-triggers triage.\n\nThe same emoji set re-triggers triage from `boucle:needs-info` (e.g. after\nthe user adds an API key following a no-key or quota-exhausted help message).\n\n### 4.3 MR approval — native forge button\n\nTo approve an MR (issue at `boucle:approval`), the human clicks the forge-native\n**Approve** button on the MR. This fires the `merge_request approved` webhook\n(dispatch.sh:167-185) → `chain_to_role merger`. A 👍 emoji on the MR does NOT\ntrigger the merger (emoji approval is spec-only).\n\n**Native-approval race** (reviewer.sh:487-497): if the human approves the MR\n*before* the reviewer finishes (issue still at `boucle:review`), the `approved`\nwebhook is silently dropped (dispatch requires `boucle:approval`\u002F`human`). The\nreviewer's PASS branch checks `forge_mr_approvals` and triggers the merger\ndirectly to recover.\n\n### 4.4 Push to `boucle\u002F\u003Ciid>-\u003Cslug>` — re-review\n\nA push to the worker branch (`boucle\u002F\u003Ciid>-\u003Cslug>`) fires `merge_request update`\n(dispatch.sh:187-205). If the issue was at `boucle:approval`, it reverts to\n`boucle:review` and re-runs the reviewer (the approval is invalidated by the\npush). Otherwise it re-runs the reviewer on the updated MR.\n\n### 4.5 MR close — label-dependent\n\nClosing the MR (not merging) fires `merge_request close` (dispatch.sh:207-258):\n- `boucle:done` \u002F `boucle:human` → no-op (terminal).\n- `boucle:approval` → escalate to `boucle:human` (user decision).\n- any other (`boucle:todo`\u002F`working`\u002F`review`\u002F`merging`) → revert to\n  `boucle:todo` + `chain_to_role worker` (fresh start).\n\n### 4.5b MR events on GitHub — vocabulary translation\n\n§4.3–4.5 describe the MR webhooks in the **GitLab** vocabulary, which is what\nthe router's `case` arms match: `open`, `update`, `close`, `reopen`,\n`approved`, `unapproved`, `merge`. GitHub sends different words, and its\npayload puts the MR under `.pull_request` rather than `.object_attributes`.\nBoth are normalized in dispatch before routing — `dispatch_github_mr_action`\nfor the verb, a two-shape jq filter for the branch and IID:\n\n| GitHub event + action | GitLab action | Routes to |\n|---|---|---|\n| `pull_request` `synchronize` | `update` | reviewer (§4.4) |\n| `pull_request` `closed`, `.pull_request.merged == true` | `merge` | catchup |\n| `pull_request` `closed`, not merged | `close` | §4.5 |\n| `pull_request` `opened` \u002F `reopened` | `open` \u002F `reopen` | §4.4 |\n| `pull_request_review` `submitted`, state `approved` | `approved` | merger (§4.3) |\n| `pull_request_review` `dismissed` | `unapproved` | reviewer |\n| any other review (commented, changes requested) | *(none)* | stays a note (§4.6) |\n| `ready_for_review`, `review_requested`, `edited` | *(none)* | skip |\n\nA merge arrives as a close on GitHub; only `.pull_request.merged` tells the two\napart, and that bit decides catchup versus a worker re-run. `synchronize` must\nalso be in the workflow's `pull_request` trigger types — without it GitHub\nnever delivers the event, and a PR is checked once at open and never again on\nlater pushes.\n\n### 4.6 Comment on MR — worker re-run with feedback\n\nA human comment on the MR (dispatch.sh:331-359) reverts the issue to\n`boucle:todo` and re-runs the worker with `BOUCLE_ITERATION=verdicts+1` and the\nMR notes injected as `BOUCLE_REVIEWER_FEEDBACK`. This is the feedback channel\nthat feeds human amendments forward (LESSONS.yml lesson #16, #53).\n\nOn GitHub the equivalent event is an `issue_comment` on a **pull request**:\nthe payload carries `.issue.number` (the PR number) with `.issue.pull_request`\nnon-null. dispatch must detect this via `dispatch_is_github_pr_comment`\n(dispatch.sh:38-49) and route it to the same MR-note handler, resolving the\nPR's source branch via `forge_mr_get` (it is not in the payload). Without\nthis, the PR's (empty) labels would trip the dispatch ABORT gate and human PR\nfeedback could never re-trigger the worker.\n\n### 4.7 Closed-issue guard\n\nAny webhook (note\u002Femoji\u002Fupdate) on a **closed issue** is a no-op\n(dispatch.sh:385-411) — EXCEPT bot assignment, which is the explicit\n\"reopen and resume\" signal (dispatch.sh:404-412). A local harness MUST NOT\nre-trigger a closed issue by any other means. (LESSONS.yml lesson #44.)\n\n### 4.8 Anti-loop filters (apply to ALL events)\n\n| Filter | Condition | Action |\n|---|---|---|\n| boucle's own comment | note body has `\u003C!-- boucle:agent -->` marker | skip (I7) |\n| boucle's own 👀 ack | emoji event, `awardable_type == Issue`, name canonicalizes to `eyes` | skip |\n| bot-originated event | `ACTOR == BOUCLE_BOT_USERNAME` and action ≠ `merge` | skip |\n| system note | `object_attributes.system == true` | skip (AGENTS.md #34) |\n| non-boucle branch | source_branch does not start with `boucle\u002F\u003Ciid>` | skip |\n\nThe 👀 filter is not redundant with the bot-identity filter: in mono-user\nmode there is no bot account, so `ACTOR` is the human on *every* event —\nincluding boucle's own reaction. Where emoji events reach dispatch at all\n(`bin\u002Fsetup` writes `emoji_events=false`, but a pre-existing hook may have\nit on), triage's own acknowledgement would otherwise land on an issue still\nlabelled `boucle:triage` — a label that routes to triage unconditionally —\nand the loop would re-triage itself.\n\nA local harness posting a comment MUST let the stamp be applied (via\n`forge_issue_note` \u002F `forge_mr_note` or by including `\u003C!-- boucle:agent -->`\nmanually) or its comment will be treated as a human reply and re-route the\nloop. (I7.)\n\n### 4.9 Cross-role chaining\n\nEvery role transition funnels through `chain_to_role \u003Ciid> \u003Crole> [var=value]`\n(lib\u002Fboucle.sh:1030) → `forge_trigger_role` (bin\u002Fforge\u002Fgitlab.sh:503 \u002F\ngithub.sh:514), which fires a pipeline trigger with `BOUCLE_ISSUE` +\n`BOUCLE_ROLE` + extra vars. A local harness does NOT call `chain_to_role`\ndirectly — it sets the label and lets dispatch route on the next webhook, OR\nit re-assigns the bot (§4.1) to re-trigger triage.\n\n---\n\n## 5. Forge abstraction contract\n\nThe engine speaks to the forge through a thin seam (`bin\u002Fforge\u002F\u003Cforge>.sh` +\n`bin\u002Fforge\u002Fcommon.sh`). A local harness interacting with boucle SHOULD use the\nsame primitives (or their forge-native equivalents) so its writes are\nindistinguishable from the engine's.\n\n| Primitive | Signature | Effect |\n|---|---|---|\n| `forge_issue_note \u003Ciid> \u003Cmsg>` | post a comment on an issue (stamped `\u003C!-- boucle:agent -->`) | |\n| `forge_mr_note \u003Cmr_iid> \u003Cmsg>` | post a comment on an MR (stamped) | |\n| `forge_issue_labels_get \u003Ciid>` | read current labels (comma-separated) | |\n| `forge_issue_labels_set \u003Ciid> \u003Clabels>` | write labels (replaces boucle: set, preserves non-boucle) | |\n| `forge_issue_assign \u003Ciid> \u003Cuser_id>` | assign the issue | |\n| `forge_mr_diff \u003Cmr_iid>` | fetch MR diff | |\n| `forge_mr_approvals \u003Cmr_iid>` | check native approval state | |\n| `forge_mr_check_suites \u003Cmr_iid>` | fetch CI check suites | |\n| `forge_trigger_role \u003Ciid> \u003Crole> [vars]` | trigger a role pipeline | |\n| `set_boucle_label \u003Ciid> \u003Cdetail> \u003Cgross>` | write detail+gross labels idempotently + reassign + notify | |\n\n**Marker stamping is mandatory.** Every comment posted via `forge_issue_note`\nor `forge_mr_note` carries `\u003C!-- boucle:agent -->`. A harness that posts via the\nraw forge API MUST add the marker manually, or dispatch will treat the comment\nas a human reply (I7).\n\n---\n\n## 6. Branch contract\n\nThe worker branch is `boucle\u002F\u003Ciid>-\u003Cslug>` (e.g. `boucle\u002F42-fix-nav-link`).\n`boucle_branch_name` (lib\u002Fboucle.sh) derives the slug deterministically from\nthe issue title: lowercase, kebab-case, max 40 chars, truncated at a word\nboundary when possible. The name MUST stay deterministic — same issue → same\nbranch across iterations. When the title is empty or the fetch fails, the\nname falls back to the legacy bare `boucle\u002F\u003Ciid>`.\n\n`boucle\u002F\u003Ciid>` remains the **protocol key**: every MR lookup\n(`forge_mr_lookup_by_branch`) prefix-matches on it, so an issue-title edit\nthat changes the slug can never break a lookup (LESSONS.yml #70, #71).\n\nThe lifecycle:\n- **Worker** (CI, `lib\u002Fboucle-ci\u002Fworker.sh`) creates the branch from\n  `master`\u002F`main`, runs the agent, pushes, opens an MR targeting\n  `master`\u002F`main`. The agent does NOT create the branch.\n- **Reviewer** reviews the MR diff \u002F deployed preview.\n- **Merger** rebases the MR onto `master`\u002F`main` (serially, via\n  `resource_group: boucle-merge`) and merges.\n- **Catchup** closes the issue after merge.\n\nA local harness taking over an issue at `boucle:human` SHOULD work on the\nexisting worker branch — `bin\u002Fboucle pause` prints its exact name — (not a\nnew branch) to preserve the worker's commits. If the branch is stale, rebase\nonto `master`\u002F`main` before pushing. If the branch is absent, create it from\n`master`\u002F`main` under the name `bin\u002Fboucle pause` prints.\n\n**Never push to `master`\u002F`main` directly** — the merger owns that transition.\nA harness that pushes to `master`\u002F`main` bypasses the serial-merge guard (I10)\nand may produce conflicts with in-flight MRs.\n\n---\n\n## 7. Known gaps\n\n- **Dogfood suspended.** The engine repo no longer dogfoods on a consumer\n  (the origin site was split out). Dogfooding will resume via a dedicated test\n  consumer once the engine\u002Fconsumer separation is stable. Until then, the\n  73 LESSONS.yml lessons catalog the forward-looking operating principles; new classes of bugs are\n  discovered on real consumers. (CONTEXT.md §1.)\n- **Lesson numbering drift.** AGENTS.md has duplicate lesson numbers (#17,\n  #22, #23, #24, #27, #28, #29, #41, #42, #47 appear twice). The doc's own rule\n  says \"never renumber — a pruned entry leaves a gap, not a shift\", but\n  collisions are not gaps. A curation pass will resolve this; until then,\n  reference lessons by their unique first sentence, not by number alone.\n\n---\n\n## 8. Interactive mode\n\nA local harness (opencode, jcode, or any agent CLI that reads markdown) can\ntake over a boucle issue, work on it interactively, and hand it back to the\nloop. This section documents the commands, the protocol, and the conventions\nthe harness MUST respect.\n\n### 8.0 The forge-native `\u002Fboucle` command\n\nThe `\u002Fboucle` command is a **user-facing capability** typed as an issue\ncomment — a forge-native observability surface, distinct from the local\n`bin\u002Fboucle` harness below. It is read-only at MVP: **no agent invocation, no\nlabel writes, no MR scope**. It gives the human the one thing labels cannot\ncarry: the content of an agent run and the detail of loop health, in the same\nchannel the loop already posts in (I1, CONTEXT.md §7).\n\n**When to use it** — you want to see what the agent did without leaving the\nissue; you want a quick health projection of the loop.\n\n**Trigger phrases** (case-insensitive, anchored at the first non-empty line):\n\n| Trigger | Verb | Returns |\n|---|---|---|\n| `\u002Fboucle log [role]` | `log` | Tail of `agent-output.log` of the most recent run of `\u003Crole>` (default: the role in flight, else the last completed) |\n| `\u002Fboucle status` | `status` | `bin\u002Fhealth` projection (iterations, outcomes by role, cost total, last verdict SHA, role in flight) |\n| `\u002Fboucle help` | `help` | the verb list + the non-redundancy rationale |\n| `@\u003CBOUCLE_BOT_USERNAME> log` | `log` | same as `\u002Fboucle log` (equivalent trigger form) |\n\n**What it does NOT do** — no label changes (labels are control\u002Fstate; `\u002Fboucle`\nis observability — never duplicate label control, I2), no agent invocation at\nMVP, no MR scope (issue comments only at MVP).\n\n**Authorization** — `log`\u002F`status`\u002F`help`: actor ∈ {issue author, parent-issue\nhuman author via `resolve_reporter_id`, one generation}. These are\nobservability of data the actor could already see in the CI UI — no new trust\nboundary. The future `jc` verb is **maintainer-only**. System notes filtered\n(lesson #34); closed-issue guard applies (except `help`, lesson #44).\n\n**Forge asymmetries** — GitHub `log` is **post-completion only** (GitHub\nActions has no streaming log API); the future `tail` live-log verb is\n**GitLab-only**. The post-completion `log` works on both forges.\n\n**Non-redundancy note** — use labels for control; use `\u002Fboucle` for\nobservability. NEVER use `\u002Fboucle` to write a label, and NEVER use a label to\ncarry what `\u002Fboucle` observes.\n\n### Commands\n\nThe `bin\u002Fboucle` wrapper exposes eight verbs. Forge identity (project,\nhost, forge type) is auto-detected from the `origin` git remote when CI\nvars are absent — see §Forge auth below.\n\n| Command | Action |\n|---|---|\n| `bin\u002Fboucle help` | Show usage with one-line description per command and examples. Also triggered by `-h`, `--help`, or no argument. |\n| `bin\u002Fboucle list [state]` | List open boucle issues from the forge, grouped by boucle state. Optional filter: `list working`, `list human`, `list review`… Alias: `ls`. |\n| `bin\u002Fboucle status \u003Ciid>` | Synthetic view: title, forge label, assignee, MR status, iteration count, `state.md` content, last 3 iterations. Works even without local `.boucle-state\u002F\u003Ciid>\u002F`. |\n| `bin\u002Fboucle check \u003Ciid>` | Pre-work gate check: file-gate (overlap with in-flight issues), dependency gate, sibling gate. Consultation only — informs, does not block. |\n| `bin\u002Fboucle log \u003Ciid>` | Prints `.boucle-state\u002F\u003Ciid>\u002Fagent-output.log` (last run trace: tool calls, file reads, git operations). |\n| `bin\u002Fboucle pause \u003Ciid>` | Human takes over. Sets `boucle:human`, saves previous state, posts a comment. |\n| `bin\u002Fboucle resume \u003Ciid>` | Loop resumes. If the worker branch (`boucle\u002F\u003Ciid>-\u003Cslug>`) has commits ahead → `boucle:review` (reviewer + merger). Otherwise restores the previous label. |\n| `bin\u002Fboucle restart \u003Ciid>` | Loop takes the issue from scratch (`boucle:todo`). Does not restore previous state. |\n\n### Posting comments\n\nAlways use `bin\u002Fforge-note` (not raw `glab`\u002F`gh` note commands) so the\n`\u003C!-- boucle:agent -->` marker is applied. Without the marker, dispatch\ntreats the comment as a human reply and re-routes the loop (I7).\n\n```\nbin\u002Fforge-note issue \u003Ciid> --message \"comment body\"\nbin\u002Fforge-note issue \u003Ciid> --message-file \u002Fpath\u002Fto\u002Fmessage.md\nbin\u002Fforge-note issue \u003Ciid> --message-stdin \u003C file\nbin\u002Fforge-note mr \u003Cmr_iid> --message \"comment body\"\n```\n\n### Read-only inspection\n\n```\nbin\u002Fhealth \u003Ciid>        # per-issue health summary (iterations, outcomes, cost, verdict)\nbin\u002Fdoctor --audit      # read-only board audit (~20 checks)\n```\n\n### Excluded commands (CI-context only)\n\nThese `bin\u002F*` commands require CI env vars and are NOT available locally.\nUse the local alternatives:\n\n| Excluded | Role | Local alternative |\n|---|---|---|\n| `bin\u002Fdescribe-images` | Describe images via vision model | The agent CLI reads the image directly with its `Read` tool (opencode supports vision) |\n| `bin\u002Ffetch-issue-attachments` | Download issue uploads | `glab api projects\u002F:id\u002Fissues\u002F\u003Ciid>\u002Fnotes` then `glab api ...\u002Fuploads\u002F:secret\u002F:filename` (or `gh api ...`) |\n| `bin\u002Ffetch-mr-attachments` | Download MR uploads | `glab api projects\u002F:id\u002Fmerge_requests\u002F\u003Ciid>\u002Fnotes` + download |\n| `bin\u002Fbuild-evidence-pack` | Extract charter docs + diff brief | The agent CLI reads `AGENTS.md`, `CONTEXT.md`, `LOOP.md`, `SKILL.md` directly with `Read` |\n| `bin\u002Frender-preview` | Render preview HTML → PNG | The agent CLI uses the browser tool (opencode browser) or screenshot |\n| `bin\u002Fcollapse-duplicate-notes` | Dedupe bot notes | Not needed in interactive mode (the human controls what they post) |\n\n### Forge auth in local mode\n\nSet these env vars before running `bin\u002Fboucle` or `bin\u002Fforge-note`:\n\n```bash\nexport BOUCLE_BOT_TOKEN=\"glpat-...\"   # or GH_TOKEN for GitHub\nexport BOUCLE_FORGE_HOST=\"gitlab.com\" # or github.com\nexport BOUCLE_PROJECT_PATH=\"org\u002Frepo\" # project path\nexport BOUCLE_FORGE=\"gitlab\"          # or \"github\"\n```\n\nThe agent CLI can also use `glab`\u002F`gh` directly for forge operations (list\nMRs, view issues, etc.) — `bin\u002Fforge-note` is just a wrapper that applies the\n`\u003C!-- boucle:agent -->` stamp.\n\n### Message templates\n\nThe `templates\u002F` directory at the repo root contains one `.md` file per\nboucle message type, with `{{placeholders}}`. These are the single source\nof truth for the format. To produce a boucle-valid message:\n\n1. Read the template (e.g. `templates\u002Ftriage.md`).\n2. Fill the placeholders with the actual content.\n3. Post via `bin\u002Fforge-note issue \u003Ciid> --message-file \u003Cpath>`.\n\n| Template | Purpose |\n|---|---|\n| `templates\u002Ftriage.md` | Triage comment (marker + TL;DR, Diagram, Analysis, one `## Criteria` section — acceptance, must-haves, non-goals —, Questions, and one collapsed `## Metadata` section: impacts, impacted files, size, validation, disposition) |\n| `templates\u002Fverdict-reviewer.md` | Reviewer verdict (marker + VERDICT:) |\n| `templates\u002Fverdict-e2e.md` | E2e verdict (marker + VERDICT:) |\n| `templates\u002Fstate.md` | state.md seed (Goal, Approach, Tried and rejected, Awaiting human) |\n| `templates\u002Fiterations.md` | iterations.md seed |\n| `templates\u002Finteractive-pause.md` | Comment posted by `boucle pause` |\n| `templates\u002Finteractive-resume.md` | Comment posted by `boucle resume` |\n\n### Protocol\n\n```\n# 1. Check the issue state\nboucle status \u003Ciid>\n\n# 2. Take over\nboucle pause \u003Ciid>\nboucle check \u003Ciid>\n\n# 3. Work in the agent CLI on the worker branch boucle\u002F\u003Ciid>-\u003Cslug>\n#    (read state.md, fill templates\u002F, commit, push)\n\n# 4. If new spec\u002Fconception was done:\n#    fill templates\u002Ftriage.md, post via bin\u002Fforge-note\n\n# 5. Hand back to the loop\nboucle resume \u003Ciid>   # if code pushed → review+merge\nboucle restart \u003Ciid>  # if starting from scratch\n```\n\n### Critical rules\n\n1. **Work on the worker branch `boucle\u002F\u003Ciid>-\u003Cslug>`** (the name\n   `boucle pause` prints), never push to `master`\u002F`main` — the merger\n   owns that transition (I10).\n2. **Always use `bin\u002Fforge-note`** for comments — the `\u003C!-- boucle:agent -->`\n   stamp prevents dispatch from treating your comment as a human reply (I7).\n3. **Run `boucle check`** before coding to avoid file-gate conflicts with\n   in-flight issues.\n4. **Use `templates\u002F`** for any boucle-format message (triage, verdict,\n   state.md) — do not improvise the format.\n5. **Read `state.md` and `iterations.md`** before starting work — they\n   contain the goal, acceptance criteria, and what previous iterations\n   tried.\n6. **Update `state.md`** (Approach section) and **append to\n   `iterations.md`** after working — the loop's reviewer reads them.\n7. **All code, templates, and CLI output are in English** — this matches\n   the boucle convention (charter docs, agent prompts, CI scripts are all\n   in English).\n\n---\n\n## 9. Failure mode framework\n\nThe invariants (§1) are the rules; this section names the *patterns* those\nrules prevent. The LESSONS.yml lessons are incident catalogs that instantiate\nthese patterns — the framework gives them a shared vocabulary so a new failure\ncan be classified against a known class, not rediscovered from scratch.\n\nThe patterns below are adapted from the Multi-Agent Systems Architect body of\nknowledge (distilled pipeline-design patterns that independently converge on\nboucle's hard-won lessons). The mapping is boucle-specific: each pattern names\nthe detection signal, the recovery path, and the lessons that instantiate it.\n\n### 9.1 Failure taxonomy\n\nEvery agent run in the loop can fail in one of seven modes. Each mode has a\ndetection signal and a recovery path already encoded in the engine.\n\n| Failure mode | Description | Boucle detection | Boucle recovery | Lessons |\n|---|---|---|---|---|\n| **Hard** | Agent crashes, times out, or the provider is down | Empty agent log \u002F no activity traces (`is_api_down`, exit 4) | Fallback provider → human escalation | #29, #30, #32 |\n| **Silent** | Agent produces output but it is wrong, hallucinated, or a placeholder | Reviewer verdict FAIL; log-scraping fallback finds no verdict | Re-trigger worker with feedback; escalate on iteration cap | #1, #2, #5, #46 |\n| **Partial** | Agent exhausts step budget mid-draft (comment\u002Fverdict not posted) | No posted note with final marker; draft marker found in stdout | Log-scraping fallback promotes draft to final | #1, #27, #45, #47 |\n| **Contradiction** | Reviewer PASSes but human amendments are unaddressed | Reviewer grades against frozen criteria, ignores amendments | Feed amendments forward; reviewer MUST verify each amendment | #16, #53 |\n| **Cascade** | One role's bad output poisons downstream (stale preview, orphaned commit) | Preview SHA mismatch; empty MR (`commits_status`) | Re-trigger worker from preserved state; NEVER `reset --hard` a committed branch | #21, #22, #51 |\n| **Loop** | Rebase-conflict re-trigger never converges; doctor re-triggers a stuck issue | `\u003C!-- boucle:conflict-retry N -->` count; iteration cap | Bounded retry → hand conflict to agent → escalate | #43, #61, #63 |\n| **Context** | Agent ignores instructions (context overload, lost state across iterations) | `state.md` re-seeded; `iterations.md` absent | Persist `.boucle-state\u002F` to `$BOUCLE_STATE_CACHE`; restore at startup | #28 |\n\nA new incident that does not fit one of these seven modes is a new class of\nfailure — the worker MUST add a lesson AND a taxonomy entry. A new incident\nthat does fit a known mode is an instance, not a new class — the worker adds a\nlesson cross-referencing the mode, not a new taxonomy entry.\n\n### 9.2 Circuit breaker\n\nThe fallback-provider logic (lesson #30) is a circuit breaker. The states:\n\n| State | Boucle behavior | Transition |\n|---|---|---|\n| **CLOSED** (normal) | Primary provider runs the agent. Track failure via `is_api_down` \u002F `is_quota_exhausted`. | ≥1 failure (empty log, quota error) → **OPEN** |\n| **OPEN** (failing) | Do not call the primary. If `BOUCLE_FALLBACK_PROVIDER` is set, try the fallback model. If fallback also fails → human escalation. | Fallback succeeds → **HALF-OPEN**; fallback fails → human |\n| **HALF-OPEN** (testing recovery) | Fallback run completed. Next run attempts the primary again. | Primary succeeds → **CLOSED**; primary fails → **OPEN** |\n\nThe exit-4 path (`bin\u002Fjc` exits 4 on empty log \u002F no activity) is the trip\nmechanism. The diagnostic comment posted on the issue names the model and\ntells the human to check API status — this is the \"always produce something\"\nrule (§I4) applied to the failure path itself.\n\n### 9.3 Fallback chain\n\nEvery role in the loop has a fallback chain. The system MUST always produce\n*something* — a structured degraded response beats a silent failure (§I4).\n\n| Priority | Agent | Condition | Boucle implementation |\n|---|---|---|---|\n| 1 (primary) | Full-capability model (e.g. deepseek-v4.1-flash) | Default | `AGENT` env var per role |\n| 2 (fallback) | Fallback model on exit-4 | Primary down \u002F quota exhausted | `BOUCLE_FALLBACK_PROVIDER` + `BOUCLE_FALLBACK_MODEL_\u003CROLE>` (lesson #30) |\n| 3 (degraded) | Log-scraping fallback (post from stdout) | Agent exhausted steps before posting | `bin\u002Fjc` scrapes `agent-output.log` for drafted verdict\u002Ftriage (lessons #27, #47) |\n| 4 (human) | Human escalation | All automated paths fail | `boucle:human` + diagnostic comment (lesson #59) |\n\n### 9.4 HITL gate calibration\n\n`boucle:human` is the HITL gate. The calibration problem applies directly:\n\n- **Over-escalation** → the human rubber-stamps → the gate is theater. Caused\n  by: iteration cap too low, reviewer too strict, doctor too aggressive.\n- **Under-escalation** → the human never sees edge cases → false confidence.\n  Caused by: reviewer PASSing unaddressed amendments (lesson #53), doctor not\n  scanning `boucle:merging` (lesson #68).\n\nGate placement in boucle is label-driven, not code-driven:\n\n| Gate | Label | Criterion | Type |\n|---|---|---|---|\n| Spec review | `boucle:spec-review` | Triage validated spec (Size S) | Blocking approval (human replies\u002Femoji) |\n| MR approval | `boucle:approval` | Reviewer PASSed | Blocking approval (native Approve button) |\n| Escalation | `boucle:human` | Iteration cap \u002F UNCERTAIN \u002F destructive | Blocking review (human acts) |\n| DND skip | `boucle:dnd` (transient) | DND window active | Auto-skip with trace (lesson #47) |\n| Autonomous skip | `boucle:autonomous` (transient) | Per-issue opt-in | Auto-skip with trace (lesson #47) |\n\n### 9.5 Observability contract\n\nThe engine's observability is `agent-output.log` + the marker\u002Fverdict system.\nThe minimum per-run record:\n\n| Field | Boucle source | How to access |\n|---|---|---|\n| trace_id | `BOUCLE_ISSUE` + `BOUCLE_ROLE` + pipeline ID | CI variables |\n| agent_id | `$AGENT` (model name) | env var |\n| latency | job duration | CI job timing |\n| input\u002Foutput tokens | agent stdout (if logged) | `agent-output.log` |\n| output | posted comment \u002F verdict | forge API |\n| confidence | verdict (PASS \u002F FAIL \u002F UNCERTAIN) | `VERDICT:` line |\n| status | success \u002F failure \u002F partial \u002F escalated | exit code + label transition |\n| errors | agent stderr + `bin\u002Fjc` diagnostics | `agent-output.log` |\n\nThe root-cause analysis protocol maps to the doctor's audit: identify blast\nradius (which issue\u002FMR), trace backward (which role produced the bad output),\nisolate the failure (agent vs upstream vs inter-role contract), classify\n(prompt ambiguity \u002F context overload \u002F model limitation \u002F schema mismatch \u002F\nmissing info), fix + regression test.\n\n### 9.6 Context budget\n\nThe loop's context management is `state.md` + `iterations.md` persisted to\n`$BOUCLE_STATE_CACHE` (lesson #28). This is the \"structured state object\"\npattern: each role reads only its required fields (goal, approach, tried,\nawaiting) and writes only its output fields. The checkpoint is the\n`iterations.md` append — a compressed record of what was tried, so the next\niteration does not re-discover it.\n\n---\n\n## See also\n\n- [AGENTS.md](AGENTS.md) — contribution conventions + lessons learned (incident catalog)\n- [ARCHITECTURE.md](ARCHITECTURE.md) — engine implementation (code structure, forge adapters, CI stages)\n- [CONTEXT.md](CONTEXT.md) — identity, audience, philosophy, constraints\n- [README.md](README.md) — overview, getting started\n- [.jcode\u002FUPSTREAM-FIX-WORKFLOW.md](.jcode\u002FUPSTREAM-FIX-WORKFLOW.md) — upstream fix workflow\n","# SKILL.md — Boucle protocol\n\n> **Source of truth for the boucle protocol.** This file is the behavioral\n> contract: the labels, markers, state machine, and handoff rules that boucle\n> speaks. It is runtime-agnostic (does not reference jcode internals) and\n> forge-agnostic (covers GitLab + GitHub). A local harness (opencode, jcode,\n> pi, or any agent runtime that reads markdown) loads this skill to \"speak\n> boucle\" — interact with a boucle loop on any consumer repo correctly.\n>\n> **Projection at install:** `bin\u002Fsetup` symlinks `.jcode\u002Fskills\u002Fboucle\u002FSKILL.md`\n> → `..\u002F..\u002FSKILL.md` so the consumer's harness loads it from the conventional\n> skills directory. The source is this file; the symlink is the projection.\n>\n> **Enforcement:** `bin\u002Fcheck-doc-sync` validates that the labels and markers\n> in the engine code match this file. A label or marker in code that is absent\n> here fails CI red — this file is the spec, not a description.\n>\n> **Relationship to other charters:**\n> - `AGENTS.md` — contribution conventions + mandatory principles. `LESSONS.yml` — lessons (incident\n>   catalog). Lessons that state a current protocol invariant cross-reference\n>   this file (`See SKILL.md §\u003Cid>`); the normative text lives here.\n> - `ARCHITECTURE.md` — engine implementation (code structure, forge adapters,\n>   CI stages). The *how*, not the *what*.\n> - `CONTEXT.md` — identity, audience, philosophy, constraints. The *why*.\n> - `README.md` — overview, getting started. The *entry*.\n\n---\n\n## 1. Invariants\n\nThe protocol rests on ten invariants. Every agent (CI or local harness) and\nevery human interacting with boucle MUST honor them. The LESSONS.yml lessons are\nincident catalogs that instantiate these invariants; the normative statement\nlives here.\n\n### I1 — Forge-native\n\nBoucle lives in the forge. NEVER introduce a new frontend, a server, or a\ncomputer to keep running. The forge is the UI; labels are the state; comments\nare the channel. (CONTEXT.md §7, LESSONS.yml lesson #55.)\n\n### I2 — Label-driven state machine\n\nLabels are the source of truth. No external database, no separate state API.\nThe state machine is driven by `boucle:*` labels (detail axis) + `boucle::status::*`\nlabels (gross axis: `bot`\u002F`human`\u002F`done`). A label change IS the state\ntransition. (CONTEXT.md §7.)\n\n### I3 — Async by design\n\nThe human is not in front of the screen. Work happens asynchronously, driven\nby labels and comments. The loop does not wait for a human reply — it progresses\nto the next gate and parks at a human-readable state (`boucle:spec-review`,\n`boucle:approval`, `boucle:human`). (CONTEXT.md §1.)\n\n### I4 — Post-early\n\nPost the comment or verdict FIRST, then refine. An incomplete draft posted is\nALWAYS better than a refinement never posted. Step-budget waste (the agent\nexhausts its budget without posting) is bug #1. (LESSONS.yml lesson #1, #2, #5.)\n\n**\"Post early\" means minimal but meaningful, not an empty placeholder.** A\ndraft MUST contain at least the content the human needs to act on it — for\ntriage, a rough `## Analysis` section (2-3 sentences restating the issue) and\na `## Metadata` section carrying a `Disposition`. An empty placeholder (\"DRAFT — first-pass triage,\nrefining next.\") is noise the human cannot act on: no analysis, no questions,\nno criteria. The agent MUST read the issue body (already in its prompt as\n`$BOUCLE_ISSUE_BODY`) before posting — \"post early\" does NOT mean \"post\nbefore reading the issue\". (LESSONS.yml lesson #99.)\n\n### I5 — Idempotence\n\nAll `bin\u002F*` scripts and label writes MUST be idempotent. Re-running a script\nproduces no additional side effects. A label PUT that does not change the label\nset is skipped (the forge records a Resource Label Event on every PUT, even a\nno-op — CONTEXT.md §8). (LESSONS.yml lesson #4.)\n\n### I6 — SHA-anchored verdicts\n\nReviewer and e2e verdicts MUST include the commit SHA as bare hex: no quotes, no\nwhitespace, no angle brackets. Exact format:\n`\u003C!-- boucle:verdict v=1 role=reviewer sha=abc123def456 -->`. The CI parser\nFAILS if the format is not respected. (LESSONS.yml lesson #6, #41, #47.)\n\n### I7 — Marker-based self-recognition\n\nBoucle recognizes its own writes by an invisible stamp\n(`\u003C!-- boucle:agent -->`), NEVER by the actor's identity. A comment posted\nwithout the stamp is treated as a human reply and routed. This is mono-user-safe\n(when bot and human share an account, the marker discriminates; the actor does\nnot). Every comment a harness posts MUST carry the stamp, or dispatch will treat\nit as a human reply and re-route. (LESSONS.yml lesson #55.)\n\nThe same marker discriminates note authorship in prompts: CI MUST tag every\nnote injected into an agent prompt `[\u003Cauthor> — boucle]` when its body carries\nthe stamp, `[\u003Cauthor> — human]` otherwise, and agents MUST classify authorship\nby that tag — NEVER by the posting account (in mono-user mode the account is\nshared, so identity discriminates nothing; a human amendment posted under the\nbot's account is still a binding amendment, lesson #111). (LESSONS.yml lesson\n#111.)\n\n### I8 — Doc-as-code\n\nA doc that describes a system that no longer exists is a bug. Charter docs are\nmaintained as part of each work cycle: triage identifies impacted docs, worker\nupdates them in the same MR, reviewer verifies conformance, e2e verifies\nproduction match. `bin\u002Fcheck-doc-sync` enforces code ↔ SKILL.md sync in CI.\n(AGENTS.md \"Documentation self-maintenance\".)\n\n### I9 — Upstream-first\n\nFix upstream (in boucle) FIRST, then update the consumer, then remediate\nexisting data. NEVER patch a consumer to work around a boucle defect. NEVER\nintroduce a local workaround that won't be reported upstream. (CONTEXT.md §7,\n`.jcode\u002FUPSTREAM-FIX-WORKFLOW.md`.)\n\n### I10 — Serial merge\n\nMerges are serialized via `resource_group: boucle-merge`. Each rebase is\nagainst a `master`\u002F`main` that includes previously-merged MRs. NEVER\nparallelize merges — a concurrent rebase against a stale branch produces\nconflicts and race conditions. (CONTEXT.md §7, LESSONS.yml lesson #8.)\n\n---\n\n## 2. State machine\n\nThe state machine is two-axis: a **detail** label (`boucle:\u003Cstate>`) that\nanswers \"what is the loop doing?\" and a **gross** label\n(`boucle::status::\u003Cowner>`) that answers \"whose side is this on?\". In\nmono-user mode (`BOUCLE_MONO_USER=true`) the gross axis is dropped (one actor\nowns both sides — the question is meaningless).\n\n### 2.1 Detail labels\n\n| Label | Meaning | Owner (gross) |\n|---|---|---|\n| `boucle:triage` | Awaiting triage analysis | bot |\n| `boucle:needs-info` | Triage needs more info from the reporter | human |\n| `boucle:spec-review` | Spec validated by triage, awaiting human spec approval | human |\n| `boucle:approved` | Human approved the spec (GitHub: label added by author) | bot |\n| `boucle:todo` | Spec approved, queued for the worker | bot |\n| `boucle:working` | Worker is running | bot |\n| `boucle:review` | Worker shipped, awaiting reviewer verdict | bot |\n| `boucle:approval` | Reviewer PASSed, MR ready, awaiting human MR approval | human |\n| `boucle:merging` | Merger is running (rebase + merge) | bot |\n| `boucle:done` | Loop complete (merged + e2e PASS) | done |\n| `boucle:human` | Escalated to a human (iteration cap, unclear criteria, destructive change, etc.) | human |\n| `boucle:blocked` | Waiting on a dependency (sub-issue not closed) | bot |\n| `boucle:split` | Parent issue split into sub-issues, waiting for them | bot |\n| `boucle:dnd` | Transient flag: spec gate auto-validated during DND window | (rides along) |\n| `boucle:autonomous` | Transient flag: spec gate skipped per-issue opt-in | (rides along) |\n| `boucle:recurring` | Context tag: issue is part of a recurring bug class (non-blocking, survives transitions) | (rides along) |\n| `boucle:board` | The status-board issue (never dispatched) | — |\n| `boucle:scheduled` | Issue created by a schedule (cron template) | — |\n| `boucle:retro` | The retro's local per-run report issue (never dispatched) | — |\n| `boucle:report` | Upstream retro proposal — anonymized case or plan, created on `BOUCLE_UPSTREAM_REPO` (never dispatched) | — |\n\n**Removed (dead labels, pruned from `bin\u002Fsetup`):** `boucle:spec-approved`.\nMR approval uses the native forge Approve button (GitLab) or the\n`boucle:approved` label (GitHub — no native Approve button on issues).\nSpec approval uses a canonical emoji reaction (👍 ❤️ 🎉 🚀) on\n`boucle:spec-review`, the `boucle:approved` label, or the magic word\n`approved` — a text reply amends the spec and re-triggers triage, it\ndoes NOT approve.\n\n### 2.2 Gross labels\n\n| Label | Meaning |\n|---|---|\n| `boucle::status::bot` | The loop owns the next action (issue assigned to bot) |\n| `boucle::status::human` | A human owns the next action (issue assigned to human reporter) |\n| `boucle::status::done` | Terminal (loop complete) |\n\n### 2.3 State diagram\n\n```mermaid\nstateDiagram-v2\n    [*] --> triage: boucle:triage added \u002F bot assigned\n    triage --> needs_info: triage needs more info\n    triage --> spec_review: triage validated spec (Size S)\n    triage --> todo: triage validated spec (auto \u002F DND \u002F autonomous)\n    triage --> human: Size L \u002F unclear criteria \u002F destructive\n    triage --> split: issue too big, split into sub-issues\n    needs_info --> triage: human replied (note on needs-info)\n    spec_review --> todo: human approved spec (👍 ❤️ 🎉 🚀 emoji)\n    spec_review --> triage: human replied with amendment (re-triage)\n    spec_review --> human: human rejected \u002F no response\n    todo --> working: worker started\n    working --> review: worker shipped code\n    working --> todo: worker no-changes \u002F build-fail (retry, iter \u003C max)\n    working --> human: worker iteration cap \u002F API down (exit 4)\n    review --> approval: reviewer PASS\n    review --> todo: reviewer FAIL (retry, iter \u003C max)\n    review --> human: reviewer FAIL (iter cap) \u002F UNCERTAIN \u002F no verdict\n    approval --> merging: human approved MR (native Approve button)\n    approval --> review: MR updated (push to boucle\u002F\u003Ciid>-\u003Cslug>)\n    approval --> human: MR closed without merge\n    merging --> done: merge succeeded + e2e PASS\n    merging --> human: merge conflict \u002F not mergeable\n    done --> [*]\n    human --> triage: human re-assigns bot (BOT_JUST_ASSIGNED)\n    human --> [*]: human closes issue\n    split --> triage: all sub-issues closed (parent re-queued)\n    blocked --> todo: dependency closed (unblock)\n```\n\n**Entry is opt-in.** An issue opened with no `boucle:` label does NOT enter\nthe state machine: it has no boucle state at all, and the loop never touches\nit. A human hands it over by adding `boucle:triage` or by assigning the bot\n(`BOT_JUST_ASSIGNED`) — the two edges out of `[*]` above. Issues boucle opens\nfor itself (schedules, triage sub-issues, the e2e follow-up) pass\n`boucle:triage` at creation, so they enter through the same edge as anything\na human hands over. Setting `BOUCLE_ENTRY_MODE=auto` restores the legacy\nbehaviour, where the `issue opened` webhook alone was enough — appropriate\nonly for a tracker that exists for boucle and nothing else.\n\n### 2.4 Transition table\n\nThe transition table is derived from the handoff primitives in §4. Every\ntransition is effected by `set_boucle_label \u003Ciid> \u003Cdetail> \u003Cgross>` (lib\u002Fboucle.sh:626),\nwhich preserves non-boucle labels, writes the detail+gross pair idempotently,\nreassigns the issue (bot on `boucle::status::bot`, human reporter on\n`boucle::status::human`), and fires the outbound notification on the\ntransition (never on the state).\n\n---\n\n## 3. Marker reference\n\nBoucle communicates via invisible HTML-comment markers stamped on issue\u002FMR\ncomments and via structural sections in comment bodies. The markers are\nmachine-readable; the structural sections are detected by pattern. A local\nharness MUST use the markers correctly or dispatch will misroute its writes.\n\n### 3.1 Self-recognition marker\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:agent -->` | bare HTML comment | `stamp_agent_marker` (bin\u002Fforge\u002Fcommon.sh:81) on every `forge_issue_note` \u002F `forge_mr_note` | `has_agent_marker` (bin\u002Fforge\u002Fcommon.sh:92) in dispatch.sh:99-103 | Distinguishes boucle's own writes from human replies. The primary anti-loop guard (invariant I7). A harness posting a comment MUST carry this stamp or dispatch treats it as a human reply and re-routes. |\n\n### 3.2 Triage markers\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:triage v=1 -->` | `v=1` (no attrs) | triage agent (final comment); draft promoted by triage.sh:178 (`sed s\u002Fdraft role=triage\u002Ftriage v=1\u002F`) | triage.sh:163 (awk), collapse-duplicate-notes:71-72 (jq structural filter) | Marks a final triage comment. CI parser acts on it (sets disposition label, assigns, pauses). |\n| `\u003C!-- boucle:draft role=triage -->` | `role=triage` | triage agent (first-pass draft) | collapse-duplicate-notes:73 (filter), triage.sh:178 (promotion source) | Marks a draft triage comment. CI parser does NOT act on drafts; log-scraping fallback promotes to final when the agent exhausts its steps. |\n| `\u003C!-- boucle:obligations v=1 -->` | `v=1` | triage.sh:213 | (informational — marks the obligations section of a triage comment) | Marks the obligations block in a triage comment (what the human must do next). |\n| `\u003C!-- boucle:needs-info v=1 reason=no-key -->` | `v=1 reason=no-key` | triage.sh:92 (no-LLM-key gate) | triage.sh:55 (jq filter, idempotency check) | Marks a needs-info comment posted when the LLM API key is missing. Used for idempotency: triage checks if a no-key needs-info was already posted before re-posting. |\n| `\u003C!-- boucle:needs-info v=1 reason=profile -->` | `v=1 reason=profile` | lib\u002Fboucle-ci\u002Ftriage.sh (first-issue profile determination, #186) | lib\u002Fboucle-ci\u002Ftriage.sh (jq filter, idempotency check) | Marks the first-issue project-profile determination comment (detected `infra` on an auto\u002Funset install). The human confirms by 👍\u002F👎 reaction or a `profile: site\\|infra` reply; no agent run is burned while it waits. |\n| `\u003C!-- boucle:profile-applied v=1 profile=infra -->` | `v=1 profile=infra\\|site` | lib\u002Fboucle-ci\u002Ftriage.sh (profile apply path, #186) | lib\u002Fboucle-ci\u002Ftriage.sh (jq filter, idempotency check) | Marks the first-issue profile determination as applied (infra preset set, or site recorded). Once present, triage never re-posts a determination for the issue. |\n| `\u003C!-- boucle:diagram-missing v=1 kinds=\u003Ckinds> -->` | `v=1 kinds=\u003Cclosed-set>` | gates.sh:169 (spec-completeness gate) | gates.sh (re-trigger triage) | Marks a spec that declares structural impacts (Size M\u002FL) but is missing the required `## Diagram` section. CI re-triggers triage to add only the missing diagram. |\n| `\u003C!-- boucle:preview-missing v=1 kinds=\u003Ckinds> -->` | `v=1 kinds=\u003Cclosed-set>` | gates.sh:227 (spec-completeness gate) | gates.sh (re-trigger triage) | Marks a spec that declares visual impacts (Size M\u002FL) but is missing the required `preview.html` \u002F `RENDER_REQUEST`. CI re-triggers triage to add only the missing preview. |\n| `\u003C!-- boucle:diagram-invalid v=1 rounds=\u003Cn> -->` | `v=1 rounds=\u003Cn>` or `v=1 state=exhausted rounds=\u003Cn>` | gates.sh (check_diagram_syntax_gate) | gates.sh (retry-bound count) | Marks a spec whose Mermaid diagram the real parser rejects — the forge would render it as an error box. CI re-triggers triage to fix ONLY the diagram. After `BOUCLE_DIAGRAM_SYNTAX_MAX_RETRIES` rounds the marker carries `state=exhausted` and the spec goes to the human as-is instead of looping. |\n\n### 3.3 Verdict markers (reviewer \u002F e2e)\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:verdict v=1 role=reviewer sha=\u003Chex> -->` | `v=1 role=reviewer sha=\u003Cshort-sha>` | reviewer agent (final verdict); draft promoted by reviewer.sh:381 | reviewer.sh:357 (SHA-anchored), :362 (SHA-unanchored fallback) | Marks a final reviewer verdict. SHA MUST be bare hex (no quotes\u002Fwhitespace\u002Fangle brackets) — invariant I6. |\n| `\u003C!-- boucle:verdict v=1 role=e2e sha=\u003Chex> -->` | `v=1 role=e2e sha=\u003Cshort-sha>` | e2e agent (final verdict); draft promoted by e2e.sh:192 | e2e.sh:179 | Marks a final e2e verdict. Same SHA contract. |\n| `\u003C!-- boucle:draft role=reviewer -->` | `role=reviewer` | reviewer agent (first-pass draft) | collapse-duplicate-notes:79, reviewer.sh:381 (promotion source) | Draft reviewer verdict. Parser does NOT act; log-scraping promotes on step exhaustion. |\n| `\u003C!-- boucle:draft role=e2e -->` | `role=e2e` | e2e agent (first-pass draft) | collapse-duplicate-notes:86, e2e.sh:187 (promotion source) | Draft e2e verdict. Same as above. |\n\n### 3.4 Dependency \u002F hierarchy markers\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:depends-on iids=N,M -->` | `iids=\u003Ccomma-separated-IIDs>` | triage.sh:755 (on each dependent sub-issue) | bin\u002Flib\u002Fdepends-on.sh:38 (grep), dispatch.sh:427 (gate check) | Declares sub-issue dependencies. Dispatch gates the worker until all dep IIDs are closed. |\n| `\u003C!-- boucle:split-parent iids=N,M -->` | `iids=\u003Ccomma-separated-IIDs>` | triage (split operation) | catchup.sh:66, doctor.sh:746, e2e.sh:75, triage.sh:518,520 (idempotency guard) | Marks the parent issue of a split with its sub-issue IIDs. Used to cascade closure and detect duplicate splits. |\n| `\u003C!-- boucle:blocked v=1 iids=N,M -->` | `v=1 iids=\u003Ccomma-separated-IIDs>` | dispatch.sh:461, boucle.sh:1136 | (informational — marks the blocked note) | Posted on a sub-issue blocked by open siblings. The label `boucle:blocked` is the state; this marker is the note's machine-readable header. |\n| `\u003C!-- boucle:sibling-blocked v=1 sib=N -->` | `v=1 sib=\u003Cactive-sibling-IID>` | dispatch.sh:508, .gitlab-ci.yml:997 | catchup.sh:86,94, e2e.sh:95,102 (jq parse) | Posted when a sibling sub-issue is still active (working\u002Freview\u002Fapproval). The issue starts automatically once the sibling reaches done\u002Fclosed — sibling issues share the same domain and run one at a time to avoid merge conflicts. |\n| `\u003C!-- boucle:unblocked v=1 by=N -->` | `v=1 by=\u003Cclosed-dep-IID>` | catchup.sh:100, e2e.sh:109 | (informational — marks the unblock note) | Posted when a dependency closes and the worker starts. |\n| `\u003C!-- boucle:e2e-origin v=1 iid=N -->` | `v=1 iid=\u003Corigin-IID>` | e2e.sh:267 (on follow-up issues) | boucle.sh:705 (parse, to resolve the original issue from a follow-up) | Carries the original issue IID on a follow-up issue created by e2e FAIL, so the loop can cascade back. |\n| `\u003C!-- boucle:e2e-fail v=1 iid=N followup=M -->` | `v=1 iid=\u003Corigin-IID> followup=\u003Cfollowup-IID>` | e2e.sh:291 (on follow-up issue created by e2e FAIL) | (informational — links follow-up to origin) | Posted on the follow-up issue created when e2e FAILs, linking it to the original issue for cascade. |\n| `\u003C!-- boucle:e2e-escalation v=1 iid=N verdict=X -->` | `v=1 iid=\u003CIID> verdict=\u003Cuncertain\\|empty>` | e2e.sh:321,357 (on e2e UNCERTAIN or no-verdict) | (informational — structured escalation) | Posted when e2e escalates to human due to UNCERTAIN verdict or no verdict (step exhaustion). Records the verdict reason. |\n| `\u003C!-- boucle:sub-issue v=1 -->` | `v=1` | triage (split operation) | triage.sh:566 (skip in parent-body parsing) | Marks a sub-issue body. Used to avoid parsing sub-issue content as parent-issue content. |\n| `\u003C!-- boucle:recurring v=1 refs=N,M -->` | `v=1 refs=\u003Ccomma-separated-IIDs>` | triage agent (optional `## Recurring theme` section) | triage.sh (parse + label), worker.sh (inject refs as context) | Recurring-theme declaration: flags the issue as part of a recurring class and cites prior closed issues. CI applies `boucle:recurring` (context tag surviving transitions). Worker receives prior-issue summaries to find the root cause. Non-blocking: never gates. |\n| `\u003C!-- boucle:impacts v=1 kinds=... -->` | `v=1 kinds=\u003Ccomma-separated-kinds>` | triage agent (`## Metadata` section in the spec comment) | triage.sh (parse_impacts_marker), gates.sh (check_diagram_gate, check_preview_gate) | Structural\u002Fvisual impact declaration (LESSONS.yml lesson #104): the triage declares which impacts this issue has from a closed set (architecture, data-model, process, state-machine, data-flow, deployment, ui, ux, design, none). CI cross-checks: a structural kind without a `\u003C!-- boucle:diagram v=1 -->` marker → gate fails → triage re-triggered; a visual kind (ui\u002Fux\u002Fdesign) without a `preview.html` + `RENDER_REQUEST` → gate fails → triage re-triggered. Drives the deterministic diagram\u002Fpreview gate. |\n| `\u003C!-- boucle:diagram v=1 types=... -->` | `v=1 types=\u003Ccomma-separated-Mermaid-block-types>` | triage agent (`## Diagram` section in the spec comment) | triage.sh (parse_diagram_marker), gates.sh (check_diagram_gate) | Diagram declaration: lists the Mermaid block types used in the `## Diagram` section (e.g. `erDiagram,flowchart`). CI cross-checks this marker against the `Impacts` field of `## Metadata`: a structural impact declared there without a `\u003C!-- boucle:diagram v=1 -->` marker → gate fails → triage re-triggered. The Mermaid theme block (boucle.dev light\u002Ftransparent) is defined in `templates\u002Fdiagram-theme.md` — the static source of truth. |\n| `\u003C!-- boucle:diagram-missing v=1 kinds=... -->` | `v=1 kinds=\u003Ccomma-separated-kinds>` | gates.sh:169 (check_diagram_gate) | (informational — re-trigger note) | Posted when a spec declares structural impacts (`## Metadata`) but lacks the required `## Diagram` section. Triage is re-triggered to add ONLY the missing diagram — it must re-read its previous spec in the Prior discussion, not re-analyze from scratch. |\n| `\u003C!-- boucle:preview-missing v=1 kinds=... -->` | `v=1 kinds=\u003Ccomma-separated-kinds>` | gates.sh:227 (check_preview_gate) | (informational — re-trigger note) | Posted when a spec declares visual impacts (`## Metadata`) but lacks the required `preview.html` \u002F `RENDER_REQUEST`. Triage is re-triggered to add ONLY the missing preview — it must re-read its previous spec in the Prior discussion, not re-analyze from scratch. |\n| `\u003C!-- boucle:diagram-invalid v=1 rounds=... -->` | `v=1 rounds=\u003Cn>` (+ `state=exhausted` once the bound is spent) | gates.sh (check_diagram_syntax_gate) | gates.sh (counts its own markers for the retry bound) | Diagram SYNTAX verdict (LESSONS.yml lesson #109). `bin\u002Fcheck-mermaid` runs every ```mermaid fence in the spec through the real Mermaid parser; a parse error means the forge renders an error box instead of a diagram, so the gate blocks spec-review and re-triggers triage with the parser's own message. Fail-open when the parser is unavailable (no node, no modules, no network) — infrastructure never blocks the loop. |\n\n### 3.5 Operational markers\n\n| Marker | Format | Written by | Parsed by | Purpose |\n|---|---|---|---|---|\n| `\u003C!-- boucle:board v=1 -->` | `v=1` | boucle.sh:268 (status board body) | (informational — identifies the board issue) | Marks the status-board issue body. Dispatch skips `boucle:board`-labelled issues; this marker is the body's stamp. |\n| `\u003C!-- boucle:catchup v=1 iid=N state=X target=Y -->` | `v=1 iid=\u003CIID> state=\u003Cfrom> target=\u003Cto>` | catchup.sh (audit note on direct merge) | (informational — audit trail) | Posted when an MR is merged directly (not via the approval flow). Records the state at merge time for audit. The `target` field is the terminal state the loop WOULD have applied pre-e2e-on-direct-merge; the actual terminal state is decided by the e2e verdict (PASS→done, FAIL\u002FUNCERTAIN→human). |\n| `\u003C!-- boucle:commit sha=\u003Chex> -->` | `sha=\u003Cshort-sha>` | worker.sh:421-422 (in MR description + build marker) | (preview freshness assertion — worker.sh) | Anchors the MR\u002Fbuild to a commit SHA for preview freshness check (invariant I6-adjacent). |\n| `\u003C!-- boucle:diagnostic v=1 iid=N class=X trigger=Y -->` | `v=1 iid=\u003CIID> class=\u003Cfailure-class> trigger=\u003Ctrigger>` | boucle.sh:475 (escalation diagnostic) | (informational — structured escalation) | Replaces the generic \"human intervention needed\" note with a structured diagnostic (failure class + evidence + recommended action). |\n| `\u003C!-- boucle:schedule id=\u003Cname> -->` | `id=\u003Cschedule-name>` | boucle.sh:201 (on scheduled issues) | boucle.sh (dedup — last firing marker) | Marks a scheduled issue with its template name. Used to prevent duplicate firings across sweeps. |\n| `\u003C!-- boucle:state v=1 -->` | `v=1` | boucle.sh:436 (`boucle_state_save`, on worker exit) | boucle.sh:462 (`boucle_state_restore`, on a cold state cache) | Per-issue loop state (iteration log, approach, tried-and-rejected, `last-outcome`) persisted on the issue so it survives an ephemeral runner, where `BOUCLE_STATE_CACHE` does not. Collapsed, and excluded from the notes injected into agent prompts (bin\u002Fjc) so it is never re-billed as prompt input. |\n| `\u003C!-- boucle:conflict-retry N -->` | `N=\u003Cretry-count>` | boucle.sh:1099 (on re-trigger after rebase conflict) | boucle.sh:1085 (jq parse, count retries) | Posted on each re-trigger after a rebase conflict. Bounded retry count — after N retries, the conflict is handed to the worker agent for resolution. |\n| `\u003C!-- boucle:file-blocked v=1 on=N paths=... -->` | `v=1 on=\u003Cactive-issue> paths=\u003Coverlap>` | gates.sh:106 (check_file_gate — defer on file overlap) | gates.sh:192 (unblock path, last marker) | The file-impact gate (LESSONS.yml lesson #62): posted when a worker would edit files claimed by an in-flight sibling issue. The unblock path fires when the named blocker closes. |\n| `\u003C!-- boucle:files v=1 paths=... -->` | `v=1 paths=\u003Cimpacted-files>` | triage agent (embedded in the spec comment's `## Metadata` section); worker job (lib\u002Fboucle-ci\u002Fworker.sh, refresh in a separate note after the branch diff) | worker job (jq, find last marker note excluding the spec comment) | File-impact declaration (LESSONS.yml lesson #62): the triage predicts the impacted files inside its spec comment, the worker refreshes the claim in a separate machine note with the actual branch diff. Consumed by `check_file_gate` to defer parallel workers on overlap. |\n| `\u003C!-- boucle:evidence-pack v=1 -->` | `v=1` | bin\u002Fbuild-evidence-pack (header first line of the evidence pack) | — | Marks the auto-generated evidence pack (charter doc snapshot + diff brief) attached to escalations. |\n| `\u003C!-- boucle:allow-list v=1 user=\u003Cusername> -->` | `v=1 user=\u003Cauthor-username>` | boucle.sh (check_allow_list_gate — rejection note) | — | Posted on an issue whose resolved human reporter is not in `BOUCLE_ALLOWED_USERS`. The issue is not accepted by the loop (no role triggered). Fail-open when the variable is unset (legacy). |\n| `\u003C!-- boucle:interactive v=1 -->` | `v=1` | bin\u002Fboucle (cmd_pause, cmd_resume, cmd_restart — interactive harness) | — | Marks a comment posted by the interactive harness (`boucle pause`\u002F`resume`\u002F`restart`). Distinguishes a voluntary human takeover from a `boucle:human` escalation. Informational — dispatch does not act on it. |\n| `\u003C!-- boucle:approval-request v=1 -->` | `v=1` (no attrs) | reviewer agent (posted on the PR via `forge_mr_note` when `boucle_mono_user` is true) | doctor (jq: `select(.body \\| contains(\"boucle:approval-request\"))`, then `forge_note_reactions` polls for a canonical approval emoji) | Mono-user PR approval gate: GitHub blocks self-review, so the reviewer posts this note on the PR and the human reacts 👍 to approve. The doctor's `doctor_mr_approval_emoji` polls the note's reactions and triggers the merger when a non-bot user reacted with a canonical approval emoji (👍 ❤️ 🚀 🎉). Replaces the old `VERDICT: PASS` auto-merge (no human signal). |\n| `\u003C!-- boucle:opt-out v=1 -->` | `v=1` (no attrs) | dispatch.sh:929 (human label-removal webhook, #124) | dispatch.sh:926 (idempotency — marker already present), doctor.sh:499 (unlabeled scan skip) | Posted when a human removes the `boucle:*` labels (GitLab `.changes.labels.previous` \u002F GitHub `unlabeled` event): the explicit \"leave the loop\" signal. Idempotent — later events no-op against it. The doctor's unlabeled scan skips tombstones so the issue is never re-appropriated; re-adding `boucle:triage` or assigning the bot re-enters the loop normally. |\n\n### 3.6 Markers NOT in code (do not emit)\n\n| Marker | Status | Note |\n|---|---|---|\n| *(none currently)* | — | Keep this table empty — a marker that is documented here but emitted by code FAILs `bin\u002Fcheck-doc-sync`. |\n\n### 3.7 Structural signals (not HTML comments)\n\n| Signal | Format | Parsed by | Purpose |\n|---|---|---|---|\n| `## TL;DR` + `## Metadata` | two section headers in the same comment | collapse-duplicate-notes:71-72, triage.sh (PRE_RUN_TRIAGE_ID \u002F COMMENT filters) | Structural final-comment detection for triage. A draft only has `## Metadata`; a final has both `## TL;DR` and `## Metadata`. Defense-in-depth: even if the agent uses the wrong marker, the parser won't act on a draft that lacks `## TL;DR`. Specs posted before the section merge carry a standalone `## Disposition` instead of `## Metadata` — the filters accept either. |\n| `## Parent issue\\n#N` | section header + issue reference | boucle.sh:699,768,774 (awk) | Resolves the parent IID from a sub-issue body. Used by `resolve_reporter_id`, `fetch-issue-attachments`, `maybe_close_parent`. |\n| `## Depends on` | section header | bin\u002Flib\u002Fdepends-on.sh:57,61 (awk fallback) | Fallback parsing of dependencies when the `\u003C!-- boucle:depends-on -->` marker is absent. Mirrors the `## Parent issue` pattern. |\n| `## Approach` | section header in `state.md` | worker.sh:482 (sed extraction), :551 (MR description) | The worker's implementation approach, extracted from `state.md` into the MR description. The reviewer reads it to verify doc conformance. |\n| `VERDICT: PASS\\|FAIL\\|UNCERTAIN` | line-anchored (`^VERDICT:`) | reviewer.sh:299,320,358,363; e2e.sh (same pattern) | The verdict line. MUST be start-of-line anchored in greps (LESSONS.yml lesson #41) — an unanchored grep matches shell traces containing the substring. |\n| `BOT_JUST_ASSIGNED` | assignee-change detection | dispatch.sh:504-539 | The real \"re-queue after boucle:human\" mechanism (§4.1). Detected from `.changes.assignees` on the `issue update` webhook, not from a comment. |\n| Emoji `thumbsup` `heart` `rocket` `tada` | emoji award on a Note | dispatch.sh (`BOUCLE_SPEC_APPROVAL_EMOJIS=\"thumbsup heart rocket tada\"`) | Spec approval + needs-info re-trigger. Valid on an issue at `boucle:spec-review` (approve spec) or `boucle:needs-info` (re-trigger after action). MR approval is the native Approve button, NOT an emoji. |\n| Emoji `eyes` (👀) | emoji award on an **Issue** | written by triage.sh (`ack_issue_taken`, `BOUCLE_ACK_EMOJI=\"eyes\"`); discarded by dispatch.sh | **Bot → human** acknowledgement: triage has picked the issue up. The only signal boucle writes as a reaction; deliberately outside the spec-approval set so it can never read as approval. Carries no routing meaning in either direction — dispatch skips issue-level 👀 events (§4.8). |\n\n---\n\n## 4. Handoff protocol\n\nThe handoff protocol defines how a human (via the forge UI or a local harness)\ninteracts with the loop. Every handoff is a webhook event routed by dispatch.\nThe primitives below are extracted from `lib\u002Fboucle-ci\u002Fdispatch.sh` — they are\nthe real mechanisms, not invented verbs.\n\n### 4.1 Re-queue after `boucle:human` — bot reassignment\n\n**`boucle:human` is a dead-end for comments.** A plain note on an issue at\n`boucle:human` matches no dispatch branch and falls through to `exit 0` —\nnothing happens. The ONLY re-queue path is **re-assigning the issue to the bot**.\n\nWhen a human assigns the bot to an issue (any idle state: `boucle:human`,\n`boucle:needs-info` without reply, unlabeled, `boucle:todo`, `boucle:spec-review`),\nthe `issue update` webhook fires with an assignee change. Dispatch detects\n`BOT_JUST_ASSIGNED` (dispatch.sh:504-539) and routes to triage from any idle\nstate. This is the explicit \"reopen and resume\" signal.\n\n**A local harness that wants to re-queue a `boucle:human` issue MUST\nre-assign the bot, not post a comment.** (LESSONS.yml lesson #11, #44.)\n\n### 4.2 Spec approval — emoji only; a reply amends\n\nTo approve a spec (issue at `boucle:spec-review`), the human awards a\n**`thumbsup` `heart` `rocket` or `tada` emoji** on the triage\u002Fspec-review\nnote (dispatch.sh, gated by `BOUCLE_SPEC_APPROVAL_EMOJIS=\"thumbsup heart\nrocket tada\"`). The emoji triggers `chain_to_role worker`. An emoji by the\nbot account is skipped — only human emoji count.\n\nA **text reply is NOT an approval** — it is a correction\u002Famendment. A reply\nre-triggers **triage** (`chain_to_role triage`), which re-reads the human's\nnotes and posts an **updated spec** back at `boucle:spec-review` for another\napproval round. The triage validation message says so explicitly: *“A reply\nnever approves the spec.”* A bot-authored note (carrying\n`\u003C!-- boucle:agent -->`) is skipped before routing.\n\nThe doctor's orphaned-`boucle:spec-review` recovery mirrors this contract:\nan emoji re-triggers the worker; a reply re-triggers triage.\n\nThe same emoji set re-triggers triage from `boucle:needs-info` (e.g. after\nthe user adds an API key following a no-key or quota-exhausted help message).\n\n### 4.3 MR approval — native forge button\n\nTo approve an MR (issue at `boucle:approval`), the human clicks the forge-native\n**Approve** button on the MR. This fires the `merge_request approved` webhook\n(dispatch.sh:167-185) → `chain_to_role merger`. A 👍 emoji on the MR does NOT\ntrigger the merger (emoji approval is spec-only).\n\n**Native-approval race** (reviewer.sh:487-497): if the human approves the MR\n*before* the reviewer finishes (issue still at `boucle:review`), the `approved`\nwebhook is silently dropped (dispatch requires `boucle:approval`\u002F`human`). The\nreviewer's PASS branch checks `forge_mr_approvals` and triggers the merger\ndirectly to recover.\n\n### 4.4 Push to `boucle\u002F\u003Ciid>-\u003Cslug>` — re-review\n\nA push to the worker branch (`boucle\u002F\u003Ciid>-\u003Cslug>`) fires `merge_request update`\n(dispatch.sh:187-205). If the issue was at `boucle:approval`, it reverts to\n`boucle:review` and re-runs the reviewer (the approval is invalidated by the\npush). Otherwise it re-runs the reviewer on the updated MR.\n\n### 4.5 MR close — label-dependent\n\nClosing the MR (not merging) fires `merge_request close` (dispatch.sh:207-258):\n- `boucle:done` \u002F `boucle:human` → no-op (terminal).\n- `boucle:approval` → escalate to `boucle:human` (user decision).\n- any other (`boucle:todo`\u002F`working`\u002F`review`\u002F`merging`) → revert to\n  `boucle:todo` + `chain_to_role worker` (fresh start).\n\n### 4.5b MR events on GitHub — vocabulary translation\n\n§4.3–4.5 describe the MR webhooks in the **GitLab** vocabulary, which is what\nthe router's `case` arms match: `open`, `update`, `close`, `reopen`,\n`approved`, `unapproved`, `merge`. GitHub sends different words, and its\npayload puts the MR under `.pull_request` rather than `.object_attributes`.\nBoth are normalized in dispatch before routing — `dispatch_github_mr_action`\nfor the verb, a two-shape jq filter for the branch and IID:\n\n| GitHub event + action | GitLab action | Routes to |\n|---|---|---|\n| `pull_request` `synchronize` | `update` | reviewer (§4.4) |\n| `pull_request` `closed`, `.pull_request.merged == true` | `merge` | catchup |\n| `pull_request` `closed`, not merged | `close` | §4.5 |\n| `pull_request` `opened` \u002F `reopened` | `open` \u002F `reopen` | §4.4 |\n| `pull_request_review` `submitted`, state `approved` | `approved` | merger (§4.3) |\n| `pull_request_review` `dismissed` | `unapproved` | reviewer |\n| any other review (commented, changes requested) | *(none)* | stays a note (§4.6) |\n| `ready_for_review`, `review_requested`, `edited` | *(none)* | skip |\n\nA merge arrives as a close on GitHub; only `.pull_request.merged` tells the two\napart, and that bit decides catchup versus a worker re-run. `synchronize` must\nalso be in the workflow's `pull_request` trigger types — without it GitHub\nnever delivers the event, and a PR is checked once at open and never again on\nlater pushes.\n\n### 4.6 Comment on MR — worker re-run with feedback\n\nA human comment on the MR (dispatch.sh:331-359) reverts the issue to\n`boucle:todo` and re-runs the worker with `BOUCLE_ITERATION=verdicts+1` and the\nMR notes injected as `BOUCLE_REVIEWER_FEEDBACK`. This is the feedback channel\nthat feeds human amendments forward (LESSONS.yml lesson #16, #53).\n\nOn GitHub the equivalent event is an `issue_comment` on a **pull request**:\nthe payload carries `.issue.number` (the PR number) with `.issue.pull_request`\nnon-null. dispatch must detect this via `dispatch_is_github_pr_comment`\n(dispatch.sh:38-49) and route it to the same MR-note handler, resolving the\nPR's source branch via `forge_mr_get` (it is not in the payload). Without\nthis, the PR's (empty) labels would trip the dispatch ABORT gate and human PR\nfeedback could never re-trigger the worker.\n\n### 4.7 Closed-issue guard\n\nAny webhook (note\u002Femoji\u002Fupdate) on a **closed issue** is a no-op\n(dispatch.sh:385-411) — EXCEPT bot assignment, which is the explicit\n\"reopen and resume\" signal (dispatch.sh:404-412). A local harness MUST NOT\nre-trigger a closed issue by any other means. (LESSONS.yml lesson #44.)\n\n### 4.8 Anti-loop filters (apply to ALL events)\n\n| Filter | Condition | Action |\n|---|---|---|\n| boucle's own comment | note body has `\u003C!-- boucle:agent -->` marker | skip (I7) |\n| boucle's own 👀 ack | emoji event, `awardable_type == Issue`, name canonicalizes to `eyes` | skip |\n| bot-originated event | `ACTOR == BOUCLE_BOT_USERNAME` and action ≠ `merge` | skip |\n| system note | `object_attributes.system == true` | skip (AGENTS.md #34) |\n| non-boucle branch | source_branch does not start with `boucle\u002F\u003Ciid>` | skip |\n\nThe 👀 filter is not redundant with the bot-identity filter: in mono-user\nmode there is no bot account, so `ACTOR` is the human on *every* event —\nincluding boucle's own reaction. Where emoji events reach dispatch at all\n(`bin\u002Fsetup` writes `emoji_events=false`, but a pre-existing hook may have\nit on), triage's own acknowledgement would otherwise land on an issue still\nlabelled `boucle:triage` — a label that routes to triage unconditionally —\nand the loop would re-triage itself.\n\nA local harness posting a comment MUST let the stamp be applied (via\n`forge_issue_note` \u002F `forge_mr_note` or by including `\u003C!-- boucle:agent -->`\nmanually) or its comment will be treated as a human reply and re-route the\nloop. (I7.)\n\n### 4.9 Cross-role chaining\n\nEvery role transition funnels through `chain_to_role \u003Ciid> \u003Crole> [var=value]`\n(lib\u002Fboucle.sh:1030) → `forge_trigger_role` (bin\u002Fforge\u002Fgitlab.sh:503 \u002F\ngithub.sh:514), which fires a pipeline trigger with `BOUCLE_ISSUE` +\n`BOUCLE_ROLE` + extra vars. A local harness does NOT call `chain_to_role`\ndirectly — it sets the label and lets dispatch route on the next webhook, OR\nit re-assigns the bot (§4.1) to re-trigger triage.\n\n---\n\n## 5. Forge abstraction contract\n\nThe engine speaks to the forge through a thin seam (`bin\u002Fforge\u002F\u003Cforge>.sh` +\n`bin\u002Fforge\u002Fcommon.sh`). A local harness interacting with boucle SHOULD use the\nsame primitives (or their forge-native equivalents) so its writes are\nindistinguishable from the engine's.\n\n| Primitive | Signature | Effect |\n|---|---|---|\n| `forge_issue_note \u003Ciid> \u003Cmsg>` | post a comment on an issue (stamped `\u003C!-- boucle:agent -->`) | |\n| `forge_mr_note \u003Cmr_iid> \u003Cmsg>` | post a comment on an MR (stamped) | |\n| `forge_issue_labels_get \u003Ciid>` | read current labels (comma-separated) | |\n| `forge_issue_labels_set \u003Ciid> \u003Clabels>` | write labels (replaces boucle: set, preserves non-boucle) | |\n| `forge_issue_assign \u003Ciid> \u003Cuser_id>` | assign the issue | |\n| `forge_mr_diff \u003Cmr_iid>` | fetch MR diff | |\n| `forge_mr_approvals \u003Cmr_iid>` | check native approval state | |\n| `forge_mr_check_suites \u003Cmr_iid>` | fetch CI check suites | |\n| `forge_trigger_role \u003Ciid> \u003Crole> [vars]` | trigger a role pipeline | |\n| `set_boucle_label \u003Ciid> \u003Cdetail> \u003Cgross>` | write detail+gross labels idempotently + reassign + notify | |\n\n**Marker stamping is mandatory.** Every comment posted via `forge_issue_note`\nor `forge_mr_note` carries `\u003C!-- boucle:agent -->`. A harness that posts via the\nraw forge API MUST add the marker manually, or dispatch will treat the comment\nas a human reply (I7).\n\n---\n\n## 6. Branch contract\n\nThe worker branch is `boucle\u002F\u003Ciid>-\u003Cslug>` (e.g. `boucle\u002F42-fix-nav-link`).\n`boucle_branch_name` (lib\u002Fboucle.sh) derives the slug deterministically from\nthe issue title: lowercase, kebab-case, max 40 chars, truncated at a word\nboundary when possible. The name MUST stay deterministic — same issue → same\nbranch across iterations. When the title is empty or the fetch fails, the\nname falls back to the legacy bare `boucle\u002F\u003Ciid>`.\n\n`boucle\u002F\u003Ciid>` remains the **protocol key**: every MR lookup\n(`forge_mr_lookup_by_branch`) prefix-matches on it, so an issue-title edit\nthat changes the slug can never break a lookup (LESSONS.yml #70, #71).\n\nThe lifecycle:\n- **Worker** (CI, `lib\u002Fboucle-ci\u002Fworker.sh`) creates the branch from\n  `master`\u002F`main`, runs the agent, pushes, opens an MR targeting\n  `master`\u002F`main`. The agent does NOT create the branch.\n- **Reviewer** reviews the MR diff \u002F deployed preview.\n- **Merger** rebases the MR onto `master`\u002F`main` (serially, via\n  `resource_group: boucle-merge`) and merges.\n- **Catchup** closes the issue after merge.\n\nA local harness taking over an issue at `boucle:human` SHOULD work on the\nexisting worker branch — `bin\u002Fboucle pause` prints its exact name — (not a\nnew branch) to preserve the worker's commits. If the branch is stale, rebase\nonto `master`\u002F`main` before pushing. If the branch is absent, create it from\n`master`\u002F`main` under the name `bin\u002Fboucle pause` prints.\n\n**Never push to `master`\u002F`main` directly** — the merger owns that transition.\nA harness that pushes to `master`\u002F`main` bypasses the serial-merge guard (I10)\nand may produce conflicts with in-flight MRs.\n\n---\n\n## 7. Known gaps\n\n- **Dogfood suspended.** The engine repo no longer dogfoods on a consumer\n  (the origin site was split out). Dogfooding will resume via a dedicated test\n  consumer once the engine\u002Fconsumer separation is stable. Until then, the\n  73 LESSONS.yml lessons catalog the forward-looking operating principles; new classes of bugs are\n  discovered on real consumers. (CONTEXT.md §1.)\n- **Lesson numbering drift.** AGENTS.md has duplicate lesson numbers (#17,\n  #22, #23, #24, #27, #28, #29, #41, #42, #47 appear twice). The doc's own rule\n  says \"never renumber — a pruned entry leaves a gap, not a shift\", but\n  collisions are not gaps. A curation pass will resolve this; until then,\n  reference lessons by their unique first sentence, not by number alone.\n\n---\n\n## 8. Interactive mode\n\nA local harness (opencode, jcode, or any agent CLI that reads markdown) can\ntake over a boucle issue, work on it interactively, and hand it back to the\nloop. This section documents the commands, the protocol, and the conventions\nthe harness MUST respect.\n\n### 8.0 The forge-native `\u002Fboucle` command\n\nThe `\u002Fboucle` command is a **user-facing capability** typed as an issue\ncomment — a forge-native observability surface, distinct from the local\n`bin\u002Fboucle` harness below. It is read-only at MVP: **no agent invocation, no\nlabel writes, no MR scope**. It gives the human the one thing labels cannot\ncarry: the content of an agent run and the detail of loop health, in the same\nchannel the loop already posts in (I1, CONTEXT.md §7).\n\n**When to use it** — you want to see what the agent did without leaving the\nissue; you want a quick health projection of the loop.\n\n**Trigger phrases** (case-insensitive, anchored at the first non-empty line):\n\n| Trigger | Verb | Returns |\n|---|---|---|\n| `\u002Fboucle log [role]` | `log` | Tail of `agent-output.log` of the most recent run of `\u003Crole>` (default: the role in flight, else the last completed) |\n| `\u002Fboucle status` | `status` | `bin\u002Fhealth` projection (iterations, outcomes by role, cost total, last verdict SHA, role in flight) |\n| `\u002Fboucle help` | `help` | the verb list + the non-redundancy rationale |\n| `@\u003CBOUCLE_BOT_USERNAME> log` | `log` | same as `\u002Fboucle log` (equivalent trigger form) |\n\n**What it does NOT do** — no label changes (labels are control\u002Fstate; `\u002Fboucle`\nis observability — never duplicate label control, I2), no agent invocation at\nMVP, no MR scope (issue comments only at MVP).\n\n**Authorization** — `log`\u002F`status`\u002F`help`: actor ∈ {issue author, parent-issue\nhuman author via `resolve_reporter_id`, one generation}. These are\nobservability of data the actor could already see in the CI UI — no new trust\nboundary. The future `jc` verb is **maintainer-only**. System notes filtered\n(lesson #34); closed-issue guard applies (except `help`, lesson #44).\n\n**Forge asymmetries** — GitHub `log` is **post-completion only** (GitHub\nActions has no streaming log API); the future `tail` live-log verb is\n**GitLab-only**. The post-completion `log` works on both forges.\n\n**Non-redundancy note** — use labels for control; use `\u002Fboucle` for\nobservability. NEVER use `\u002Fboucle` to write a label, and NEVER use a label to\ncarry what `\u002Fboucle` observes.\n\n### Commands\n\nThe `bin\u002Fboucle` wrapper exposes eight verbs. Forge identity (project,\nhost, forge type) is auto-detected from the `origin` git remote when CI\nvars are absent — see §Forge auth below.\n\n| Command | Action |\n|---|---|\n| `bin\u002Fboucle help` | Show usage with one-line description per command and examples. Also triggered by `-h`, `--help`, or no argument. |\n| `bin\u002Fboucle list [state]` | List open boucle issues from the forge, grouped by boucle state. Optional filter: `list working`, `list human`, `list review`… Alias: `ls`. |\n| `bin\u002Fboucle status \u003Ciid>` | Synthetic view: title, forge label, assignee, MR status, iteration count, `state.md` content, last 3 iterations. Works even without local `.boucle-state\u002F\u003Ciid>\u002F`. |\n| `bin\u002Fboucle check \u003Ciid>` | Pre-work gate check: file-gate (overlap with in-flight issues), dependency gate, sibling gate. Consultation only — informs, does not block. |\n| `bin\u002Fboucle log \u003Ciid>` | Prints `.boucle-state\u002F\u003Ciid>\u002Fagent-output.log` (last run trace: tool calls, file reads, git operations). |\n| `bin\u002Fboucle pause \u003Ciid>` | Human takes over. Sets `boucle:human`, saves previous state, posts a comment. |\n| `bin\u002Fboucle resume \u003Ciid>` | Loop resumes. If the worker branch (`boucle\u002F\u003Ciid>-\u003Cslug>`) has commits ahead → `boucle:review` (reviewer + merger). Otherwise restores the previous label. |\n| `bin\u002Fboucle restart \u003Ciid>` | Loop takes the issue from scratch (`boucle:todo`). Does not restore previous state. |\n\n### Posting comments\n\nAlways use `bin\u002Fforge-note` (not raw `glab`\u002F`gh` note commands) so the\n`\u003C!-- boucle:agent -->` marker is applied. Without the marker, dispatch\ntreats the comment as a human reply and re-routes the loop (I7).\n\n```\nbin\u002Fforge-note issue \u003Ciid> --message \"comment body\"\nbin\u002Fforge-note issue \u003Ciid> --message-file \u002Fpath\u002Fto\u002Fmessage.md\nbin\u002Fforge-note issue \u003Ciid> --message-stdin \u003C file\nbin\u002Fforge-note mr \u003Cmr_iid> --message \"comment body\"\n```\n\n### Read-only inspection\n\n```\nbin\u002Fhealth \u003Ciid>        # per-issue health summary (iterations, outcomes, cost, verdict)\nbin\u002Fdoctor --audit      # read-only board audit (~20 checks)\n```\n\n### Excluded commands (CI-context only)\n\nThese `bin\u002F*` commands require CI env vars and are NOT available locally.\nUse the local alternatives:\n\n| Excluded | Role | Local alternative |\n|---|---|---|\n| `bin\u002Fdescribe-images` | Describe images via vision model | The agent CLI reads the image directly with its `Read` tool (opencode supports vision) |\n| `bin\u002Ffetch-issue-attachments` | Download issue uploads | `glab api projects\u002F:id\u002Fissues\u002F\u003Ciid>\u002Fnotes` then `glab api ...\u002Fuploads\u002F:secret\u002F:filename` (or `gh api ...`) |\n| `bin\u002Ffetch-mr-attachments` | Download MR uploads | `glab api projects\u002F:id\u002Fmerge_requests\u002F\u003Ciid>\u002Fnotes` + download |\n| `bin\u002Fbuild-evidence-pack` | Extract charter docs + diff brief | The agent CLI reads `AGENTS.md`, `CONTEXT.md`, `LOOP.md`, `SKILL.md` directly with `Read` |\n| `bin\u002Frender-preview` | Render preview HTML → PNG | The agent CLI uses the browser tool (opencode browser) or screenshot |\n| `bin\u002Fcollapse-duplicate-notes` | Dedupe bot notes | Not needed in interactive mode (the human controls what they post) |\n\n### Forge auth in local mode\n\nSet these env vars before running `bin\u002Fboucle` or `bin\u002Fforge-note`:\n\n```bash\nexport BOUCLE_BOT_TOKEN=\"glpat-...\"   # or GH_TOKEN for GitHub\nexport BOUCLE_FORGE_HOST=\"gitlab.com\" # or github.com\nexport BOUCLE_PROJECT_PATH=\"org\u002Frepo\" # project path\nexport BOUCLE_FORGE=\"gitlab\"          # or \"github\"\n```\n\nThe agent CLI can also use `glab`\u002F`gh` directly for forge operations (list\nMRs, view issues, etc.) — `bin\u002Fforge-note` is just a wrapper that applies the\n`\u003C!-- boucle:agent -->` stamp.\n\n### Message templates\n\nThe `templates\u002F` directory at the repo root contains one `.md` file per\nboucle message type, with `{{placeholders}}`. These are the single source\nof truth for the format. To produce a boucle-valid message:\n\n1. Read the template (e.g. `templates\u002Ftriage.md`).\n2. Fill the placeholders with the actual content.\n3. Post via `bin\u002Fforge-note issue \u003Ciid> --message-file \u003Cpath>`.\n\n| Template | Purpose |\n|---|---|\n| `templates\u002Ftriage.md` | Triage comment (marker + TL;DR, Diagram, Analysis, one `## Criteria` section — acceptance, must-haves, non-goals —, Questions, and one collapsed `## Metadata` section: impacts, impacted files, size, validation, disposition) |\n| `templates\u002Fverdict-reviewer.md` | Reviewer verdict (marker + VERDICT:) |\n| `templates\u002Fverdict-e2e.md` | E2e verdict (marker + VERDICT:) |\n| `templates\u002Fstate.md` | state.md seed (Goal, Approach, Tried and rejected, Awaiting human) |\n| `templates\u002Fiterations.md` | iterations.md seed |\n| `templates\u002Finteractive-pause.md` | Comment posted by `boucle pause` |\n| `templates\u002Finteractive-resume.md` | Comment posted by `boucle resume` |\n\n### Protocol\n\n```\n# 1. Check the issue state\nboucle status \u003Ciid>\n\n# 2. Take over\nboucle pause \u003Ciid>\nboucle check \u003Ciid>\n\n# 3. Work in the agent CLI on the worker branch boucle\u002F\u003Ciid>-\u003Cslug>\n#    (read state.md, fill templates\u002F, commit, push)\n\n# 4. If new spec\u002Fconception was done:\n#    fill templates\u002Ftriage.md, post via bin\u002Fforge-note\n\n# 5. Hand back to the loop\nboucle resume \u003Ciid>   # if code pushed → review+merge\nboucle restart \u003Ciid>  # if starting from scratch\n```\n\n### Critical rules\n\n1. **Work on the worker branch `boucle\u002F\u003Ciid>-\u003Cslug>`** (the name\n   `boucle pause` prints), never push to `master`\u002F`main` — the merger\n   owns that transition (I10).\n2. **Always use `bin\u002Fforge-note`** for comments — the `\u003C!-- boucle:agent -->`\n   stamp prevents dispatch from treating your comment as a human reply (I7).\n3. **Run `boucle check`** before coding to avoid file-gate conflicts with\n   in-flight issues.\n4. **Use `templates\u002F`** for any boucle-format message (triage, verdict,\n   state.md) — do not improvise the format.\n5. **Read `state.md` and `iterations.md`** before starting work — they\n   contain the goal, acceptance criteria, and what previous iterations\n   tried.\n6. **Update `state.md`** (Approach section) and **append to\n   `iterations.md`** after working — the loop's reviewer reads them.\n7. **All code, templates, and CLI output are in English** — this matches\n   the boucle convention (charter docs, agent prompts, CI scripts are all\n   in English).\n\n---\n\n## 9. Failure mode framework\n\nThe invariants (§1) are the rules; this section names the *patterns* those\nrules prevent. The LESSONS.yml lessons are incident catalogs that instantiate\nthese patterns — the framework gives them a shared vocabulary so a new failure\ncan be classified against a known class, not rediscovered from scratch.\n\nThe patterns below are adapted from the Multi-Agent Systems Architect body of\nknowledge (distilled pipeline-design patterns that independently converge on\nboucle's hard-won lessons). The mapping is boucle-specific: each pattern names\nthe detection signal, the recovery path, and the lessons that instantiate it.\n\n### 9.1 Failure taxonomy\n\nEvery agent run in the loop can fail in one of seven modes. Each mode has a\ndetection signal and a recovery path already encoded in the engine.\n\n| Failure mode | Description | Boucle detection | Boucle recovery | Lessons |\n|---|---|---|---|---|\n| **Hard** | Agent crashes, times out, or the provider is down | Empty agent log \u002F no activity traces (`is_api_down`, exit 4) | Fallback provider → human escalation | #29, #30, #32 |\n| **Silent** | Agent produces output but it is wrong, hallucinated, or a placeholder | Reviewer verdict FAIL; log-scraping fallback finds no verdict | Re-trigger worker with feedback; escalate on iteration cap | #1, #2, #5, #46 |\n| **Partial** | Agent exhausts step budget mid-draft (comment\u002Fverdict not posted) | No posted note with final marker; draft marker found in stdout | Log-scraping fallback promotes draft to final | #1, #27, #45, #47 |\n| **Contradiction** | Reviewer PASSes but human amendments are unaddressed | Reviewer grades against frozen criteria, ignores amendments | Feed amendments forward; reviewer MUST verify each amendment | #16, #53 |\n| **Cascade** | One role's bad output poisons downstream (stale preview, orphaned commit) | Preview SHA mismatch; empty MR (`commits_status`) | Re-trigger worker from preserved state; NEVER `reset --hard` a committed branch | #21, #22, #51 |\n| **Loop** | Rebase-conflict re-trigger never converges; doctor re-triggers a stuck issue | `\u003C!-- boucle:conflict-retry N -->` count; iteration cap | Bounded retry → hand conflict to agent → escalate | #43, #61, #63 |\n| **Context** | Agent ignores instructions (context overload, lost state across iterations) | `state.md` re-seeded; `iterations.md` absent | Persist `.boucle-state\u002F` to `$BOUCLE_STATE_CACHE`; restore at startup | #28 |\n\nA new incident that does not fit one of these seven modes is a new class of\nfailure — the worker MUST add a lesson AND a taxonomy entry. A new incident\nthat does fit a known mode is an instance, not a new class — the worker adds a\nlesson cross-referencing the mode, not a new taxonomy entry.\n\n### 9.2 Circuit breaker\n\nThe fallback-provider logic (lesson #30) is a circuit breaker. The states:\n\n| State | Boucle behavior | Transition |\n|---|---|---|\n| **CLOSED** (normal) | Primary provider runs the agent. Track failure via `is_api_down` \u002F `is_quota_exhausted`. | ≥1 failure (empty log, quota error) → **OPEN** |\n| **OPEN** (failing) | Do not call the primary. If `BOUCLE_FALLBACK_PROVIDER` is set, try the fallback model. If fallback also fails → human escalation. | Fallback succeeds → **HALF-OPEN**; fallback fails → human |\n| **HALF-OPEN** (testing recovery) | Fallback run completed. Next run attempts the primary again. | Primary succeeds → **CLOSED**; primary fails → **OPEN** |\n\nThe exit-4 path (`bin\u002Fjc` exits 4 on empty log \u002F no activity) is the trip\nmechanism. The diagnostic comment posted on the issue names the model and\ntells the human to check API status — this is the \"always produce something\"\nrule (§I4) applied to the failure path itself.\n\n### 9.3 Fallback chain\n\nEvery role in the loop has a fallback chain. The system MUST always produce\n*something* — a structured degraded response beats a silent failure (§I4).\n\n| Priority | Agent | Condition | Boucle implementation |\n|---|---|---|---|\n| 1 (primary) | Full-capability model (e.g. deepseek-v4.1-flash) | Default | `AGENT` env var per role |\n| 2 (fallback) | Fallback model on exit-4 | Primary down \u002F quota exhausted | `BOUCLE_FALLBACK_PROVIDER` + `BOUCLE_FALLBACK_MODEL_\u003CROLE>` (lesson #30) |\n| 3 (degraded) | Log-scraping fallback (post from stdout) | Agent exhausted steps before posting | `bin\u002Fjc` scrapes `agent-output.log` for drafted verdict\u002Ftriage (lessons #27, #47) |\n| 4 (human) | Human escalation | All automated paths fail | `boucle:human` + diagnostic comment (lesson #59) |\n\n### 9.4 HITL gate calibration\n\n`boucle:human` is the HITL gate. The calibration problem applies directly:\n\n- **Over-escalation** → the human rubber-stamps → the gate is theater. Caused\n  by: iteration cap too low, reviewer too strict, doctor too aggressive.\n- **Under-escalation** → the human never sees edge cases → false confidence.\n  Caused by: reviewer PASSing unaddressed amendments (lesson #53), doctor not\n  scanning `boucle:merging` (lesson #68).\n\nGate placement in boucle is label-driven, not code-driven:\n\n| Gate | Label | Criterion | Type |\n|---|---|---|---|\n| Spec review | `boucle:spec-review` | Triage validated spec (Size S) | Blocking approval (human replies\u002Femoji) |\n| MR approval | `boucle:approval` | Reviewer PASSed | Blocking approval (native Approve button) |\n| Escalation | `boucle:human` | Iteration cap \u002F UNCERTAIN \u002F destructive | Blocking review (human acts) |\n| DND skip | `boucle:dnd` (transient) | DND window active | Auto-skip with trace (lesson #47) |\n| Autonomous skip | `boucle:autonomous` (transient) | Per-issue opt-in | Auto-skip with trace (lesson #47) |\n\n### 9.5 Observability contract\n\nThe engine's observability is `agent-output.log` + the marker\u002Fverdict system.\nThe minimum per-run record:\n\n| Field | Boucle source | How to access |\n|---|---|---|\n| trace_id | `BOUCLE_ISSUE` + `BOUCLE_ROLE` + pipeline ID | CI variables |\n| agent_id | `$AGENT` (model name) | env var |\n| latency | job duration | CI job timing |\n| input\u002Foutput tokens | agent stdout (if logged) | `agent-output.log` |\n| output | posted comment \u002F verdict | forge API |\n| confidence | verdict (PASS \u002F FAIL \u002F UNCERTAIN) | `VERDICT:` line |\n| status | success \u002F failure \u002F partial \u002F escalated | exit code + label transition |\n| errors | agent stderr + `bin\u002Fjc` diagnostics | `agent-output.log` |\n\nThe root-cause analysis protocol maps to the doctor's audit: identify blast\nradius (which issue\u002FMR), trace backward (which role produced the bad output),\nisolate the failure (agent vs upstream vs inter-role contract), classify\n(prompt ambiguity \u002F context overload \u002F model limitation \u002F schema mismatch \u002F\nmissing info), fix + regression test.\n\n### 9.6 Context budget\n\nThe loop's context management is `state.md` + `iterations.md` persisted to\n`$BOUCLE_STATE_CACHE` (lesson #28). This is the \"structured state object\"\npattern: each role reads only its required fields (goal, approach, tried,\nawaiting) and writes only its output fields. The checkpoint is the\n`iterations.md` append — a compressed record of what was tried, so the next\niteration does not re-discover it.\n\n---\n\n## See also\n\n- [AGENTS.md](AGENTS.md) — contribution conventions + lessons learned (incident catalog)\n- [ARCHITECTURE.md](ARCHITECTURE.md) — engine implementation (code structure, forge adapters, CI stages)\n- [CONTEXT.md](CONTEXT.md) — identity, audience, philosophy, constraints\n- [README.md](README.md) — overview, getting started\n- [.jcode\u002FUPSTREAM-FIX-WORKFLOW.md](.jcode\u002FUPSTREAM-FIX-WORKFLOW.md) — upstream fix workflow\n",{"name":5,"description":7},"main","b1e46a5b01665a6910c6cfc503880b3156429f73","",0,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle","https:\u002F\u002Fboucle.dev","https:\u002F\u002Fgithub.com\u002Fankaboot-source","https:\u002F\u002Fwww.ankaboot.io\u002F","customer-engagement tech · privacy by design","2026-11-14 05:53:11",{"source":8,"skillName":5,"commands":36,"prompt":46,"zipAvailable":47},[37,40,43],{"tool":38,"command":39},"npx","npx skills add ankaboot-source\u002Fboucle --skill boucle",{"tool":41,"command":42},"bunx","bunx skills add ankaboot-source\u002Fboucle --skill boucle",{"tool":44,"command":45},"pnpm","pnpm dlx skills add ankaboot-source\u002Fboucle --skill boucle","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FSKILL.md",true,[49,55,59,65,69,73,77,81,85,89,93,97,101,105,109,113,117,121,125,129,135,139,143,147,151,155,159,163,167,171,177,181,185,189,193,197,201,205,209,213,217,221,225,229,233,237,241,245,249,253,257,261,265,269,273,277,281,285,289,293,296,300,304,308,312,316,320,324,328,332,336,340,344,348,352,356,360,364,368,372,376,380,384,388,392,396,400,404,408,412,414,418,422,426,430,434,438,442,446,450,454,458,462,466,470,474,478,482,486,490,494,498,502,506,510,514,518,522,526,530,534,538,542,546,550,554,558,562,566,570,574,578,582,586,590,594,598,602,606,610,614,618,622,626,630,634,638,642,646,650,654,658,662,666,670,674,678,682,686,690,694,698,702,706],{"path":50,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":53,"sourceHtmlUrl":54},".boucle-version","OTHER","其他",8,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.boucle-version",{"path":56,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":57,"sourceHtmlUrl":58},".editorconfig",438,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.editorconfig",{"path":60,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":63,"sourceHtmlUrl":64},".github\u002Fdependabot.yml","REFERENCE","参考文档",147,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.github\u002Fdependabot.yml",{"path":66,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":67,"sourceHtmlUrl":68},".github\u002Fworkflows\u002Fboucle.yml",90789,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.github\u002Fworkflows\u002Fboucle.yml",{"path":70,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":71,"sourceHtmlUrl":72},".gitignore",910,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.gitignore",{"path":74,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":75,"sourceHtmlUrl":76},".gitlab-ci.yml",72669,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.gitlab-ci.yml",{"path":78,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":79,"sourceHtmlUrl":80},".gitmodules",259,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.gitmodules",{"path":82,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":83,"sourceHtmlUrl":84},".jcode\u002Fagents\u002Fe2e.md",9361,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fagents\u002Fe2e.md",{"path":86,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":87,"sourceHtmlUrl":88},".jcode\u002Fagents\u002Fretro-impl.md",3788,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fagents\u002Fretro-impl.md",{"path":90,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":91,"sourceHtmlUrl":92},".jcode\u002Fagents\u002Fretro.md",7527,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fagents\u002Fretro.md",{"path":94,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":95,"sourceHtmlUrl":96},".jcode\u002Fagents\u002Freviewer.md",25166,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fagents\u002Freviewer.md",{"path":98,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":99,"sourceHtmlUrl":100},".jcode\u002Fagents\u002Ftriage.md",64762,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fagents\u002Ftriage.md",{"path":102,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":103,"sourceHtmlUrl":104},".jcode\u002Fagents\u002Fworker.md",33388,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fagents\u002Fworker.md",{"path":106,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":107,"sourceHtmlUrl":108},".jcode\u002FDESIGN-template.md",4848,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002FDESIGN-template.md",{"path":110,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":111,"sourceHtmlUrl":112},".jcode\u002Fskills\u002Femilkowalski-NOTICE.md",224,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Femilkowalski-NOTICE.md",{"path":114,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":115,"sourceHtmlUrl":116},".jcode\u002Fskills\u002Fgsap-NOTICE.md",235,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fgsap-NOTICE.md",{"path":118,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":119,"sourceHtmlUrl":120},".jcode\u002Fskills\u002Fleonxlnx-NOTICE.md",215,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fleonxlnx-NOTICE.md",{"path":122,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":123,"sourceHtmlUrl":124},".jcode\u002Fskills\u002Fmattpocock-LICENSE.txt",1068,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fmattpocock-LICENSE.txt",{"path":126,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":127,"sourceHtmlUrl":128},".jcode\u002Fskills\u002Fmattpocock-NOTICE.md",190,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fmattpocock-NOTICE.md",{"path":130,"resourceType":131,"resourceTypeDesc":132,"sizeBytes":133,"sourceHtmlUrl":134},".jcode\u002Fskills\u002Ftrailofbits\u002FLICENSE","LICENSE","许可证",20131,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Ftrailofbits\u002FLICENSE",{"path":136,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":137,"sourceHtmlUrl":138},".jcode\u002Fskills\u002Ftrailofbits\u002FNOTICE.md",823,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Ftrailofbits\u002FNOTICE.md",{"path":140,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":141,"sourceHtmlUrl":142},".jcode\u002FUPSTREAM-FIX-WORKFLOW.md",1706,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002FUPSTREAM-FIX-WORKFLOW.md",{"path":144,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":145,"sourceHtmlUrl":146},".pre-commit-config.yaml",3193,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.pre-commit-config.yaml",{"path":148,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":149,"sourceHtmlUrl":150},".shellcheckrc",1220,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.shellcheckrc",{"path":152,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":153,"sourceHtmlUrl":154},"AGENTS.md",32118,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FAGENTS.md",{"path":156,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":157,"sourceHtmlUrl":158},"ARCHITECTURE.md",35915,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FARCHITECTURE.md",{"path":160,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":161,"sourceHtmlUrl":162},"bin\u002Fboucle",25607,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fboucle",{"path":164,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":165,"sourceHtmlUrl":166},"bin\u002Fboucle-ci",873,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fboucle-ci",{"path":168,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":169,"sourceHtmlUrl":170},"bin\u002Fbuild-evidence-pack",6288,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fbuild-evidence-pack",{"path":172,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":175,"sourceHtmlUrl":176},"bin\u002Fcheck-bootstrap.sh","SCRIPT","脚本",4288,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-bootstrap.sh",{"path":178,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":179,"sourceHtmlUrl":180},"bin\u002Fcheck-boucle-sync",4835,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-boucle-sync",{"path":182,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":183,"sourceHtmlUrl":184},"bin\u002Fcheck-design-charter",4955,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-design-charter",{"path":186,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":187,"sourceHtmlUrl":188},"bin\u002Fcheck-doc-sync",6686,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-doc-sync",{"path":190,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":191,"sourceHtmlUrl":192},"bin\u002Fcheck-lessons",14179,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-lessons",{"path":194,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":195,"sourceHtmlUrl":196},"bin\u002Fcheck-mermaid",4438,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-mermaid",{"path":198,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":199,"sourceHtmlUrl":200},"bin\u002Fcheck-shell",9697,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcheck-shell",{"path":202,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":203,"sourceHtmlUrl":204},"bin\u002Fcollapse-duplicate-notes",12252,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fcollapse-duplicate-notes",{"path":206,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":207,"sourceHtmlUrl":208},"bin\u002Fdescribe-images",14538,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fdescribe-images",{"path":210,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":211,"sourceHtmlUrl":212},"bin\u002Fdnd",3798,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fdnd",{"path":214,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":215,"sourceHtmlUrl":216},"bin\u002Fdoctor",32076,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fdoctor",{"path":218,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":219,"sourceHtmlUrl":220},"bin\u002Ffetch-issue-attachments",8447,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Ffetch-issue-attachments",{"path":222,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":223,"sourceHtmlUrl":224},"bin\u002Ffetch-mr-attachments",7762,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Ffetch-mr-attachments",{"path":226,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":227,"sourceHtmlUrl":228},"bin\u002Fforge-note",4500,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fforge-note",{"path":230,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":231,"sourceHtmlUrl":232},"bin\u002Fforge\u002Fcommon.sh",29373,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fforge\u002Fcommon.sh",{"path":234,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":235,"sourceHtmlUrl":236},"bin\u002Fforge\u002Fgithub.sh",51719,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fforge\u002Fgithub.sh",{"path":238,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":239,"sourceHtmlUrl":240},"bin\u002Fforge\u002Fgitlab.sh",38744,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fforge\u002Fgitlab.sh",{"path":242,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":243,"sourceHtmlUrl":244},"bin\u002Fhealth",4828,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fhealth",{"path":246,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":247,"sourceHtmlUrl":248},"bin\u002Fjc",145931,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fjc",{"path":250,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":251,"sourceHtmlUrl":252},"bin\u002Flib\u002Fagent-browser.sh",4511,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Flib\u002Fagent-browser.sh",{"path":254,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":255,"sourceHtmlUrl":256},"bin\u002Flib\u002Fdepends-on.sh",7502,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Flib\u002Fdepends-on.sh",{"path":258,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":259,"sourceHtmlUrl":260},"bin\u002Flib\u002Fengine-symlink.sh",4191,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Flib\u002Fengine-symlink.sh",{"path":262,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":263,"sourceHtmlUrl":264},"bin\u002Fnav-assert",8534,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fnav-assert",{"path":266,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":267,"sourceHtmlUrl":268},"bin\u002Frender-preview",9500,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Frender-preview",{"path":270,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":271,"sourceHtmlUrl":272},"bin\u002Fretro-post",3353,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fretro-post",{"path":274,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":275,"sourceHtmlUrl":276},"bin\u002Fsetup",102123,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fsetup",{"path":278,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":279,"sourceHtmlUrl":280},"bin\u002Fskills-index",4276,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fskills-index",{"path":282,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":283,"sourceHtmlUrl":284},"bin\u002Fskills-stats",14926,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fskills-stats",{"path":286,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":287,"sourceHtmlUrl":288},"bin\u002Fupdate",41625,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fbin\u002Fupdate",{"path":290,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":291,"sourceHtmlUrl":292},"CONTEXT.md",13112,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FCONTEXT.md",{"path":294,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":267,"sourceHtmlUrl":295},"docker\u002FDockerfile.agents","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fdocker\u002FDockerfile.agents",{"path":297,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":298,"sourceHtmlUrl":299},"docs\u002Fcost-benchmark.md",22258,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fdocs\u002Fcost-benchmark.md",{"path":301,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":302,"sourceHtmlUrl":303},"docs\u002Fprime-agent-study.md",47116,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fdocs\u002Fprime-agent-study.md",{"path":305,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":306,"sourceHtmlUrl":307},"docs\u002Fschedules-example.md",1282,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fdocs\u002Fschedules-example.md",{"path":309,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":310,"sourceHtmlUrl":311},"docs\u002Fskills-audit.md",28566,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fdocs\u002Fskills-audit.md",{"path":313,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":314,"sourceHtmlUrl":315},"install.sh",1327,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Finstall.sh",{"path":317,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":318,"sourceHtmlUrl":319},"LESSONS.yml",165573,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FLESSONS.yml",{"path":321,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":322,"sourceHtmlUrl":323},"lib\u002Fboucle-ci.sh",4431,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci.sh",{"path":325,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":326,"sourceHtmlUrl":327},"lib\u002Fboucle-ci\u002Fautomerge.sh",5406,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fautomerge.sh",{"path":329,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":330,"sourceHtmlUrl":331},"lib\u002Fboucle-ci\u002Fcatchup.sh",9381,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fcatchup.sh",{"path":333,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":334,"sourceHtmlUrl":335},"lib\u002Fboucle-ci\u002Fcheck.sh",1794,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fcheck.sh",{"path":337,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":338,"sourceHtmlUrl":339},"lib\u002Fboucle-ci\u002Fcmd-help.sh",1716,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fcmd-help.sh",{"path":341,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":342,"sourceHtmlUrl":343},"lib\u002Fboucle-ci\u002Fcmd-log.sh",2018,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fcmd-log.sh",{"path":345,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":346,"sourceHtmlUrl":347},"lib\u002Fboucle-ci\u002Fcmd-status.sh",932,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fcmd-status.sh",{"path":349,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":350,"sourceHtmlUrl":351},"lib\u002Fboucle-ci\u002Fcommand.sh",5793,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fcommand.sh",{"path":353,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":354,"sourceHtmlUrl":355},"lib\u002Fboucle-ci\u002Fdeploy.sh",776,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fdeploy.sh",{"path":357,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":358,"sourceHtmlUrl":359},"lib\u002Fboucle-ci\u002Fdispatch.sh",73681,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fdispatch.sh",{"path":361,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":362,"sourceHtmlUrl":363},"lib\u002Fboucle-ci\u002Fdoctor.sh",71446,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fdoctor.sh",{"path":365,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":366,"sourceHtmlUrl":367},"lib\u002Fboucle-ci\u002Fe2e.sh",24712,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fe2e.sh",{"path":369,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":370,"sourceHtmlUrl":371},"lib\u002Fboucle-ci\u002Fgates.sh",23844,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fgates.sh",{"path":373,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":374,"sourceHtmlUrl":375},"lib\u002Fboucle-ci\u002Fmerger.sh",11779,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fmerger.sh",{"path":377,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":378,"sourceHtmlUrl":379},"lib\u002Fboucle-ci\u002Fpost-merge.sh",5223,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fpost-merge.sh",{"path":381,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":382,"sourceHtmlUrl":383},"lib\u002Fboucle-ci\u002Fretro.sh",51043,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fretro.sh",{"path":385,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":386,"sourceHtmlUrl":387},"lib\u002Fboucle-ci\u002Freviewer.sh",56918,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Freviewer.sh",{"path":389,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":390,"sourceHtmlUrl":391},"lib\u002Fboucle-ci\u002Ftriage.sh",86929,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Ftriage.sh",{"path":393,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":394,"sourceHtmlUrl":395},"lib\u002Fboucle-ci\u002Fworker.sh",61691,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle-ci\u002Fworker.sh",{"path":397,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":398,"sourceHtmlUrl":399},"lib\u002Fboucle.sh",172982,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fboucle.sh",{"path":401,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":402,"sourceHtmlUrl":403},"lib\u002Ffreeride.sh",5881,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Ffreeride.sh",{"path":405,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":406,"sourceHtmlUrl":407},"lib\u002Fmermaid-parse.mjs",6385,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fmermaid-parse.mjs",{"path":409,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":410,"sourceHtmlUrl":411},"lib\u002Fprofile.sh",12469,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Flib\u002Fprofile.sh",{"path":131,"resourceType":131,"resourceTypeDesc":132,"sizeBytes":413,"sourceHtmlUrl":16},34523,{"path":415,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":416,"sourceHtmlUrl":417},"LOOP-README.md",3782,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FLOOP-README.md",{"path":419,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":420,"sourceHtmlUrl":421},"LOOP.md",119353,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FLOOP.md",{"path":423,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":424,"sourceHtmlUrl":425},"Makefile",5606,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FMakefile",{"path":427,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":428,"sourceHtmlUrl":429},"README.md",29156,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002FREADME.md",{"path":431,"resourceType":173,"resourceTypeDesc":174,"sizeBytes":432,"sourceHtmlUrl":433},"scripts\u002Ftest.sh",2300,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Fscripts\u002Ftest.sh",{"path":435,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":436,"sourceHtmlUrl":437},"templates\u002Fdiagram-theme.md",10811,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Fdiagram-theme.md",{"path":439,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":440,"sourceHtmlUrl":441},"templates\u002Finteractive-pause.md",240,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Finteractive-pause.md",{"path":443,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":444,"sourceHtmlUrl":445},"templates\u002Finteractive-resume.md",79,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Finteractive-resume.md",{"path":447,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":448,"sourceHtmlUrl":449},"templates\u002Fiterations.md",187,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Fiterations.md",{"path":451,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":452,"sourceHtmlUrl":453},"templates\u002Fstate.md",193,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Fstate.md",{"path":455,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":456,"sourceHtmlUrl":457},"templates\u002Ftriage.md",2352,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Ftriage.md",{"path":459,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":460,"sourceHtmlUrl":461},"templates\u002Fverdict-e2e.md",101,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Fverdict-e2e.md",{"path":463,"resourceType":61,"resourceTypeDesc":62,"sizeBytes":464,"sourceHtmlUrl":465},"templates\u002Fverdict-reviewer.md",106,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftemplates\u002Fverdict-reviewer.md",{"path":467,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":468,"sourceHtmlUrl":469},"test\u002Fack-pickup.bats",8509,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fack-pickup.bats",{"path":471,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":472,"sourceHtmlUrl":473},"test\u002Fautomerge.bats",23942,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fautomerge.bats",{"path":475,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":476,"sourceHtmlUrl":477},"test\u002Fbase-control.bats",3304,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fbase-control.bats",{"path":479,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":480,"sourceHtmlUrl":481},"test\u002Fboard.bats",9417,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fboard.bats",{"path":483,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":484,"sourceHtmlUrl":485},"test\u002Fboucle-lib.bats",83815,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fboucle-lib.bats",{"path":487,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":488,"sourceHtmlUrl":489},"test\u002Fcatchup.bats",1168,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcatchup.bats",{"path":491,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":492,"sourceHtmlUrl":493},"test\u002Fcheck-boucle-sync.bats",4521,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcheck-boucle-sync.bats",{"path":495,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":496,"sourceHtmlUrl":497},"test\u002Fcheck-doc-sync.bats",2661,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcheck-doc-sync.bats",{"path":499,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":500,"sourceHtmlUrl":501},"test\u002Fcheck-lessons.bats",7463,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcheck-lessons.bats",{"path":503,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":504,"sourceHtmlUrl":505},"test\u002Fcheck-shell.bats",4621,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcheck-shell.bats",{"path":507,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":508,"sourceHtmlUrl":509},"test\u002Fci-needs-dag.bats",14514,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fci-needs-dag.bats",{"path":511,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":512,"sourceHtmlUrl":513},"test\u002Fcollapse-duplicate-notes.bats",18416,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcollapse-duplicate-notes.bats",{"path":515,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":516,"sourceHtmlUrl":517},"test\u002Fcommand.bats",8855,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcommand.bats",{"path":519,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":520,"sourceHtmlUrl":521},"test\u002Fconsumer-ci-noise.bats",12941,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fconsumer-ci-noise.bats",{"path":523,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":524,"sourceHtmlUrl":525},"test\u002Fcost.bats",8637,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fcost.bats",{"path":527,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":528,"sourceHtmlUrl":529},"test\u002Fdeploy-stage.bats",9944,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdeploy-stage.bats",{"path":531,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":532,"sourceHtmlUrl":533},"test\u002Fdescribe-images.bats",3752,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdescribe-images.bats",{"path":535,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":536,"sourceHtmlUrl":537},"test\u002Fdesign-charter.bats",6702,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdesign-charter.bats",{"path":539,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":540,"sourceHtmlUrl":541},"test\u002Fdiagram-syntax.bats",14484,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdiagram-syntax.bats",{"path":543,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":544,"sourceHtmlUrl":545},"test\u002Fdispatch.bats",66267,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdispatch.bats",{"path":547,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":548,"sourceHtmlUrl":549},"test\u002Fdnd.bats",5792,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdnd.bats",{"path":551,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":552,"sourceHtmlUrl":553},"test\u002Fdoctor-audit.bats",5776,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdoctor-audit.bats",{"path":555,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":556,"sourceHtmlUrl":557},"test\u002Fdoctor-cadence.bats",4302,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdoctor-cadence.bats",{"path":559,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":560,"sourceHtmlUrl":561},"test\u002Fdoctor-on-dispatch.bats",11201,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdoctor-on-dispatch.bats",{"path":563,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":564,"sourceHtmlUrl":565},"test\u002Fdoctor.bats",21897,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fdoctor.bats",{"path":567,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":568,"sourceHtmlUrl":569},"test\u002Fe2e-command.bats",8441,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fe2e-command.bats",{"path":571,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":572,"sourceHtmlUrl":573},"test\u002Fengine-symlink.bats",6316,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fengine-symlink.bats",{"path":575,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":576,"sourceHtmlUrl":577},"test\u002Ffetch-issue-attachments.bats",6021,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Ffetch-issue-attachments.bats",{"path":579,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":580,"sourceHtmlUrl":581},"test\u002Ffile-gate.bats",13442,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Ffile-gate.bats",{"path":583,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":584,"sourceHtmlUrl":585},"test\u002Fforge-github.bats",12848,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fforge-github.bats",{"path":587,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":588,"sourceHtmlUrl":589},"test\u002Fforge-gitlab.bats",3844,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fforge-gitlab.bats",{"path":591,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":592,"sourceHtmlUrl":593},"test\u002Fforge-note.bats",7791,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fforge-note.bats",{"path":595,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":596,"sourceHtmlUrl":597},"test\u002Fforge-pagination.bats",5434,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fforge-pagination.bats",{"path":599,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":600,"sourceHtmlUrl":601},"test\u002Fforge-reactions.bats",4799,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fforge-reactions.bats",{"path":603,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":604,"sourceHtmlUrl":605},"test\u002Ffreeride.bats",5066,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Ffreeride.bats",{"path":607,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":608,"sourceHtmlUrl":609},"test\u002Fhealth.bats",11351,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fhealth.bats",{"path":611,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":612,"sourceHtmlUrl":613},"test\u002Fiteration-budget.bats",7395,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fiteration-budget.bats",{"path":615,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":616,"sourceHtmlUrl":617},"test\u002Fjc.bats",77599,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fjc.bats",{"path":619,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":620,"sourceHtmlUrl":621},"test\u002Fmerge-gate.bats",5468,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fmerge-gate.bats",{"path":623,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":624,"sourceHtmlUrl":625},"test\u002Fmetrics.bats",58410,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fmetrics.bats",{"path":627,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":628,"sourceHtmlUrl":629},"test\u002Fnav-assert.bats",7145,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fnav-assert.bats",{"path":631,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":632,"sourceHtmlUrl":633},"test\u002Fno-consumer-traces.bats",7178,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fno-consumer-traces.bats",{"path":635,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":636,"sourceHtmlUrl":637},"test\u002Fpreview-check.bats",15386,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fpreview-check.bats",{"path":639,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":640,"sourceHtmlUrl":641},"test\u002Fprofile.bats",7416,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fprofile.bats",{"path":643,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":644,"sourceHtmlUrl":645},"test\u002Fretro-stage.bats",34812,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fretro-stage.bats",{"path":647,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":648,"sourceHtmlUrl":649},"test\u002Fretro.bats",24604,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fretro.bats",{"path":651,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":652,"sourceHtmlUrl":653},"test\u002Freview-command.bats",4428,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Freview-command.bats",{"path":655,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":656,"sourceHtmlUrl":657},"test\u002Freviewer-images.bats",4184,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Freviewer-images.bats",{"path":659,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":660,"sourceHtmlUrl":661},"test\u002Frunner-toolchain.bats",7947,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Frunner-toolchain.bats",{"path":663,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":664,"sourceHtmlUrl":665},"test\u002Fschedules.bats",7962,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fschedules.bats",{"path":667,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":668,"sourceHtmlUrl":669},"test\u002Fsetup-infra-profile.bats",5649,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fsetup-infra-profile.bats",{"path":671,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":672,"sourceHtmlUrl":673},"test\u002Fsetup.bats",5712,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fsetup.bats",{"path":675,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":676,"sourceHtmlUrl":677},"test\u002Fskills.bats",8711,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fskills.bats",{"path":679,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":680,"sourceHtmlUrl":681},"test\u002Fspec-gate.bats",13031,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fspec-gate.bats",{"path":683,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":684,"sourceHtmlUrl":685},"test\u002Fspec-sections.bats",19229,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fspec-sections.bats",{"path":687,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":688,"sourceHtmlUrl":689},"test\u002Fstate-note.bats",7058,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fstate-note.bats",{"path":691,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":692,"sourceHtmlUrl":693},"test\u002Ftakeover.bats",3987,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Ftakeover.bats",{"path":695,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":696,"sourceHtmlUrl":697},"test\u002Fupdate.bats",31631,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fupdate.bats",{"path":699,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":700,"sourceHtmlUrl":701},"test\u002Fworker-build-gate.bats",4143,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fworker-build-gate.bats",{"path":703,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":704,"sourceHtmlUrl":705},"test\u002Fworker-engine-dirt.bats",5636,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fworker-engine-dirt.bats",{"path":707,"resourceType":51,"resourceTypeDesc":52,"sizeBytes":708,"sourceHtmlUrl":709},"test\u002Fworker-retry.bats",2958,"https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002Ftest\u002Fworker-retry.bats",[711,718,726,735,742,749],{"skillNo":712,"name":713,"nameZh":6,"description":714,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":715,"categories":716,"keywords":717,"publishedAt":20},"01M79PEN7B3R3TJ19FZFZHTGPW","wrangler","Cloudflare Workers CLI for deploying, developing, and managing Workers, KV, R2, D1, Vectorize, Hyperdrive, Workers AI, Containers, Queues, Workflows, Pipelines, and Secrets Store. Load before running wrangler commands to ensure correct syntax and best practices. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fwrangler\u002FSKILL.md",[],[],{"skillNo":719,"name":720,"nameZh":6,"description":721,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":722,"licenseFileUrl":6,"sourceHtmlUrl":723,"categories":724,"keywords":725,"publishedAt":20},"01M79PEE1ZRT65Z4AW0N7X9D3C","web-design-guidelines","Web platform design and accessibility guidelines. Use when building web interfaces, auditing accessibility, implementing responsive layouts, or reviewing web UI code. Triggers on tasks involving HTML, CSS, web components, WCAG compliance, responsive design, or web performance.","MIT","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fweb-design-guidelines\u002FSKILL.md",[],[],{"skillNo":727,"name":728,"nameZh":6,"description":729,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":730,"licenseFileUrl":731,"sourceHtmlUrl":732,"categories":733,"keywords":734,"publishedAt":20},"01M79PE7NFMZCVF5BQ7CNHP73Y","web-artifacts-builder","Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn\u002Fui). Use for complex artifacts requiring state management, routing, or shadcn\u002Fui components - not for simple single-file HTML\u002FJSX artifacts.","Apache-2.0","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fweb-artifacts-builder\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fweb-artifacts-builder\u002FSKILL.md",[],[],{"skillNo":736,"name":737,"nameZh":6,"description":738,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":739,"categories":740,"keywords":741,"publishedAt":20},"01M79PE4AYTMSWM8C35RQ780YK","wayfinder","Plan a huge chunk of work — more than one agent session can hold — as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fwayfinder\u002FSKILL.md",[],[],{"skillNo":743,"name":744,"nameZh":6,"description":745,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":746,"categories":747,"keywords":748,"publishedAt":20},"01M79PE17KCVNPPFRDW5BN76PW","verification-before-completion","Use when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fverification-before-completion\u002FSKILL.md",[],[],{"skillNo":750,"name":751,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":752,"categories":753,"keywords":754,"publishedAt":20},"01M79PDWHJB2ZGK7F94VYNT64Y","ui-ux-pro-max","https:\u002F\u002Fgithub.com\u002Fankaboot-source\u002Fboucle\u002Fblob\u002Fmain\u002F.jcode\u002Fskills\u002Fui-ux-pro-max\u002FSKILL.md",[],[]]