[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6CZFXDVMEW2688P1P456S0D":3,"skill-same:01M6CZFXDVMEW2688P1P456S0D":57},{"skillNo":4,"name":5,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":14,"categories":15,"keywords":16,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":12,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":6,"updatedAt":32,"install":33,"resources":46},"01M6CZFXDVMEW2688P1P456S0D","clawq-ultra-maintenance",null,"clankercode\u002Fclawq","clankercode","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F269201082?v=4","01M6CZFKG57M58N3F63P2H9742","01M6CZFKMHDHR7D6ADH22V9T11",0,"Unlicense","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002F.claude\u002Fskills\u002Fclawq-ultra-maintenance\u002FSKILL.md",[],[17,18,19],"maintenance","clawq","clawq ultra maintenance","2026-11-10 10:12:30",false,"---\nname: clawq-ultra-maintenance\ndescription: Comprehensive, file-coordinated maintenance sweep of clawq — fan-out subagents find bugs AND refactoring opportunities, triage into a grouped master backlog, then fix each group in an isolated worktree behind a review+codex gate, auto-merging to master. Use for a thorough \"ultra\" maintenance pass over the codebase (or an optional focus path). Resumable: all state lives on disk under .maint\u002F\u003Crun-id>\u002F.\n---\n\n# clawq-ultra-maintenance\n\nA maintenance workflow built by hand out of files + subagents — **no Workflow tool required**. A gitignored\n`.maint\u002F\u003Crun-id>\u002F` directory is the journal: it holds all state so the run is **resumable**, and each phase reads\nthe prior phase's files (handoff) while a claims dir gives fixers mutual exclusion (sync).\n\nYou are the **coordinator** (this main session). You drive phase transitions and maintain the ledger; subagents do\nthe per-unit work and report **by writing files**, never by messaging you.\n\n**Core principle:** parallelism = several subagent dispatches in one message (they return when all finish). Files are\ndurable state + handoff + inter-agent sync — *not* a polling channel. Treat a missing\u002Fincomplete output file exactly\nas a workflow treats a `null` agent result: re-dispatch that unit; never advance a phase on partial output.\n\n## Inputs\n\n- **`focus-path`** (optional arg): restrict scope to a path\u002Fcomponent (e.g. `src\u002Fdiscord.ml`, `src\u002F`). Default: whole\n  `src\u002F` + `test\u002F`.\n- On invoke, detect resume: if the newest `.maint\u002F*\u002Fledger.json` has `phase != \"done\"` and a matching scope, **resume\n  it** (start at the first incomplete phase). Otherwise mint a fresh `run_id` = `date +%Y%m%d-%H%M%S`.\n\n## State layout\n\n```\n.maint\u002F\u003Crun-id>\u002F\n  ledger.json          # phase + per-phase status + run config; the resume anchor (you maintain it with Write\u002FEdit)\n  partition.md         # P1: every target file assigned to exactly one finder bucket\n  findings\u002F\u003Cfinder>.md # P2: append-only; entries = file:line · bug|maint · sev(H\u002FM\u002FL) · desc · suggested direction\n  backlog.md           # P3: deduped master list, grouped into mostly-disjoint conceptual areas\n  groups\u002F\u003Cg>.plan.md   # P4: concrete fix plan per group (what \u002F where \u002F why \u002F which test suites)\n  claims\u002F\u003Cg>.lock\u002F     # P5: atomic claim dir (mkdir) → one fixer per group; survives resume\n  reports\u002F\u003Cg>.md       # P5\u002F6: files changed, tests run, gate result, ending with `=== DONE ===`\n  LOG.md               # your human-readable narrator log of phase transitions and any dropped\u002Fskipped items\n```\n\n**DONE marker:** every finder\u002Ffixer output file MUST end with a line `=== DONE ===`. A file lacking it = incomplete =\nre-dispatch.\n\n**ledger.json shape:**\n```json\n{ \"run_id\": \"...\", \"scope\": \"src\u002F test\u002F\", \"mode\": \"both\", \"concurrency\": 3,\n  \"phase\": \"find\",\n  \"phases\": {\"init\":\"done\",\"partition\":\"done\",\"find\":\"in_progress\",\"triage\":\"todo\",\"plan\":\"todo\",\"fix\":\"todo\",\"gate\":\"todo\",\"final\":\"todo\"},\n  \"finders\": [\"...\"], \"groups\": [\"...\"] }\n```\n\n## Workflow\n\nRun phases in order. After each, set its status `done` in `ledger.json` and advance `phase`. Append a one-line entry\nto `LOG.md` at every transition.\n\n### P0 — Init\n- Create `.maint\u002F\u003Crun-id>\u002F` and subdirs (`findings\u002F groups\u002F claims\u002F reports\u002F`).\n- Ensure `.maint\u002F` is gitignored (it is NOT by default) — append `.maint\u002F` to `.gitignore` and `.dockerignore` if\n  absent. (`.worktrees\u002F` is already ignored.)\n- Write the initial `ledger.json`.\n\n### P1 — Survey & partition\n- List target files (`*.ml`\u002F`*.mli` under scope, plus `test\u002F`). Bucket them by component\u002Fdirectory so **coverage is\n  complete** — every file in exactly one bucket. Scale finder count to size (~1 finder per 15–25 files or per\n  cohesive component; sensible cap ~8–10).\n- Write `partition.md` (bucket → file list) and record `finders` in the ledger.\n\n### P2 — Find (parallel)\n- Dispatch one subagent per bucket **in a single message**. Each gets the finder prompt (see\n  `references\u002Fprompt-pack.md` → *Finder*), its file list, and its output path `findings\u002F\u003Cfinder>.md`.\n- **Validate on return:** every dispatched finder's file exists and ends with `=== DONE ===`. Re-dispatch any\n  missing\u002Fincomplete before advancing. Log any finder that produced nothing.\n\n### P3 — Triage & group\n- Read all `findings\u002F*.md`. Dedupe, drop false-positives and low-value noise, keep legitimate bug + maintainability\n  items. (Large finding set → dispatch one triage subagent with the *Triage* prompt; otherwise do it yourself.)\n- Write `backlog.md`: master list, each item `file:line · bug|maint · sev · description`, organized into **mostly\n  disjoint conceptual groups** (so groups can be fixed as independent batches without colliding). Record `groups`.\n\n### P4 — Plan per group\n- For each group write `groups\u002F\u003Cg>.plan.md`: concrete plan — what changes, where (file:line), why, and the **exact\n  test suites** to run (`make test-run ARGS=\"test \u003Cregex>\"`). Respect file-size limits (split >1000 LoC files).\n- This is the handoff that informs P5.\n\n### P5 — Fix (parallel, isolated, capped)\n- Dispatch fixers in **waves of `concurrency` (default 3)** — system-load norm; each fixer builds\u002Ftests in its own\n  worktree. Give each the *Fixer* prompt, the group plan path, and `backlog.md`.\n- Each fixer: `mkdir .maint\u002F\u003Crun-id>\u002Fclaims\u002F\u003Cg>.lock` (atomic — if it fails, the group is already claimed → skip) →\n  create worktree `.worktrees\u002Fmaint-\u003Cg>` on branch `maint-\u003Crun-id>-\u003Cg>` → implement the plan → run **only the named\n  test suites** (never `make test-all`; never run `dune` in parallel within a worktree) → commit at intervals → run\n  `review-and-fix` over its own diff (use `pirfl` for hard\u002Fmulti-step items) → write `reports\u002F\u003Cg>.md` ending\n  `=== DONE ===`. The fixer does **not** merge.\n- Validate each wave's reports (DONE marker) before the next wave.\n\n### P6 — Gate + merge (sequential per group)\n- Merges must be serialized (one branch into master at a time). For each group with a DONE report:\n  1. Run an external review: invoke `ccc-review-cx` over the group's diff + `groups\u002F\u003Cg>.plan.md` + relevant\n     `backlog.md` rows. (If `ccc-review-cx` is unavailable, fall back to a second `review-and-fix` pass with the\n     reviewer model `openai-codex:gpt-5.5`; if no external reviewer at all, a self `review-and-fix` pass.)\n  2. **PASS** → merge `maint-\u003Crun-id>-\u003Cg>` into master, then remove the worktree (first inspect\n     `git status --porcelain --ignored` and rescue any untracked\u002Fgitignored scratch — never blind `rm -rf`\u002F`--force`).\n  3. **FAIL** → dispatch a follow-up fixer to address the review findings in the same worktree, then re-gate.\n- Append the gate result to `reports\u002F\u003Cg>.md` and mark the group merged in the ledger.\n\n### P7 — Final pass\n- Run `review-and-fix` over the whole merged diff. Run a `ccc-review-cx` review of `backlog.md` for **completeness**\n  (was every legitimate item actually fixed?). Fix gaps via a fixer agent; re-gate.\n- When the review PASSes, run `\u002Fcarm` to commit-all + rebase master. Write a final summary to `LOG.md` and set\n  `phase: done`.\n\n## Cross-cutting rules\n\n- **Resumability:** on every invoke, read the ledger and resume at the first incomplete phase. Claimed\u002Fmerged groups\n  are skipped (claims dir + ledger). Persist artifacts to disk *as each unit finishes*, not only at the end.\n- **Validate, never silently truncate:** missing output \u002F absent `=== DONE ===` ⇒ re-dispatch. If you intentionally\n  drop or defer anything (a finding, a group, a file), record it in `LOG.md` — no silent gaps.\n- **Subagents must NEVER use `attn`.** Put this line in every dispatch prompt. Agents report by writing their file.\n- **Dune\u002Ftests:** never two `dune` commands at once in the same checkout; run only the focused suites a change\n  touches; never `make test-all` in a subagent; if you must run broad tests, capture output to a temp file.\n- **Dependency floor:** the orchestration needs only bash + subagent dispatch. `review-and-fix`, `pirfl`,\n  `ccc-review-cx`, `carm` are invoked by agents and **degrade gracefully** if absent (fall back to self-review).\n\n## Agent prompts\n\nVerbatim, reusable prompt templates (Finder \u002F Triage \u002F Fixer \u002F Gate, plus the boilerplate every dispatch must carry)\nlive in [`references\u002Fprompt-pack.md`](references\u002Fprompt-pack.md). Design rationale and constraints are in\n[`references\u002Fmethodology.md`](references\u002Fmethodology.md).\n\nEvery dispatch prompt MUST include: the output file path + `=== DONE ===` contract; \"report by writing your file, not\nby messaging\"; \"**never use `attn`**\"; and the relevant dune\u002Ftest rules above.\n\n## Done checklist\n\n- [ ] `.maint\u002F` + `.worktrees\u002F` gitignored; ledger reaches `phase: done`.\n- [ ] Every target file covered by exactly one finder bucket (P1) and every finder validated (DONE).\n- [ ] `backlog.md` deduped and grouped; every legitimate item lands in a group.\n- [ ] Each group fixed in its own worktree, focused tests green, self-reviewed, gated (external review PASS), merged,\n      worktree removed.\n- [ ] Final `review-and-fix` + completeness review PASS; `\u002Fcarm` run; summary in `LOG.md`.\n","# clawq-ultra-maintenance\n\nA maintenance workflow built by hand out of files + subagents — **no Workflow tool required**. A gitignored\n`.maint\u002F\u003Crun-id>\u002F` directory is the journal: it holds all state so the run is **resumable**, and each phase reads\nthe prior phase's files (handoff) while a claims dir gives fixers mutual exclusion (sync).\n\nYou are the **coordinator** (this main session). You drive phase transitions and maintain the ledger; subagents do\nthe per-unit work and report **by writing files**, never by messaging you.\n\n**Core principle:** parallelism = several subagent dispatches in one message (they return when all finish). Files are\ndurable state + handoff + inter-agent sync — *not* a polling channel. Treat a missing\u002Fincomplete output file exactly\nas a workflow treats a `null` agent result: re-dispatch that unit; never advance a phase on partial output.\n\n## Inputs\n\n- **`focus-path`** (optional arg): restrict scope to a path\u002Fcomponent (e.g. `src\u002Fdiscord.ml`, `src\u002F`). Default: whole\n  `src\u002F` + `test\u002F`.\n- On invoke, detect resume: if the newest `.maint\u002F*\u002Fledger.json` has `phase != \"done\"` and a matching scope, **resume\n  it** (start at the first incomplete phase). Otherwise mint a fresh `run_id` = `date +%Y%m%d-%H%M%S`.\n\n## State layout\n\n```\n.maint\u002F\u003Crun-id>\u002F\n  ledger.json          # phase + per-phase status + run config; the resume anchor (you maintain it with Write\u002FEdit)\n  partition.md         # P1: every target file assigned to exactly one finder bucket\n  findings\u002F\u003Cfinder>.md # P2: append-only; entries = file:line · bug|maint · sev(H\u002FM\u002FL) · desc · suggested direction\n  backlog.md           # P3: deduped master list, grouped into mostly-disjoint conceptual areas\n  groups\u002F\u003Cg>.plan.md   # P4: concrete fix plan per group (what \u002F where \u002F why \u002F which test suites)\n  claims\u002F\u003Cg>.lock\u002F     # P5: atomic claim dir (mkdir) → one fixer per group; survives resume\n  reports\u002F\u003Cg>.md       # P5\u002F6: files changed, tests run, gate result, ending with `=== DONE ===`\n  LOG.md               # your human-readable narrator log of phase transitions and any dropped\u002Fskipped items\n```\n\n**DONE marker:** every finder\u002Ffixer output file MUST end with a line `=== DONE ===`. A file lacking it = incomplete =\nre-dispatch.\n\n**ledger.json shape:**\n```json\n{ \"run_id\": \"...\", \"scope\": \"src\u002F test\u002F\", \"mode\": \"both\", \"concurrency\": 3,\n  \"phase\": \"find\",\n  \"phases\": {\"init\":\"done\",\"partition\":\"done\",\"find\":\"in_progress\",\"triage\":\"todo\",\"plan\":\"todo\",\"fix\":\"todo\",\"gate\":\"todo\",\"final\":\"todo\"},\n  \"finders\": [\"...\"], \"groups\": [\"...\"] }\n```\n\n## Workflow\n\nRun phases in order. After each, set its status `done` in `ledger.json` and advance `phase`. Append a one-line entry\nto `LOG.md` at every transition.\n\n### P0 — Init\n- Create `.maint\u002F\u003Crun-id>\u002F` and subdirs (`findings\u002F groups\u002F claims\u002F reports\u002F`).\n- Ensure `.maint\u002F` is gitignored (it is NOT by default) — append `.maint\u002F` to `.gitignore` and `.dockerignore` if\n  absent. (`.worktrees\u002F` is already ignored.)\n- Write the initial `ledger.json`.\n\n### P1 — Survey & partition\n- List target files (`*.ml`\u002F`*.mli` under scope, plus `test\u002F`). Bucket them by component\u002Fdirectory so **coverage is\n  complete** — every file in exactly one bucket. Scale finder count to size (~1 finder per 15–25 files or per\n  cohesive component; sensible cap ~8–10).\n- Write `partition.md` (bucket → file list) and record `finders` in the ledger.\n\n### P2 — Find (parallel)\n- Dispatch one subagent per bucket **in a single message**. Each gets the finder prompt (see\n  `references\u002Fprompt-pack.md` → *Finder*), its file list, and its output path `findings\u002F\u003Cfinder>.md`.\n- **Validate on return:** every dispatched finder's file exists and ends with `=== DONE ===`. Re-dispatch any\n  missing\u002Fincomplete before advancing. Log any finder that produced nothing.\n\n### P3 — Triage & group\n- Read all `findings\u002F*.md`. Dedupe, drop false-positives and low-value noise, keep legitimate bug + maintainability\n  items. (Large finding set → dispatch one triage subagent with the *Triage* prompt; otherwise do it yourself.)\n- Write `backlog.md`: master list, each item `file:line · bug|maint · sev · description`, organized into **mostly\n  disjoint conceptual groups** (so groups can be fixed as independent batches without colliding). Record `groups`.\n\n### P4 — Plan per group\n- For each group write `groups\u002F\u003Cg>.plan.md`: concrete plan — what changes, where (file:line), why, and the **exact\n  test suites** to run (`make test-run ARGS=\"test \u003Cregex>\"`). Respect file-size limits (split >1000 LoC files).\n- This is the handoff that informs P5.\n\n### P5 — Fix (parallel, isolated, capped)\n- Dispatch fixers in **waves of `concurrency` (default 3)** — system-load norm; each fixer builds\u002Ftests in its own\n  worktree. Give each the *Fixer* prompt, the group plan path, and `backlog.md`.\n- Each fixer: `mkdir .maint\u002F\u003Crun-id>\u002Fclaims\u002F\u003Cg>.lock` (atomic — if it fails, the group is already claimed → skip) →\n  create worktree `.worktrees\u002Fmaint-\u003Cg>` on branch `maint-\u003Crun-id>-\u003Cg>` → implement the plan → run **only the named\n  test suites** (never `make test-all`; never run `dune` in parallel within a worktree) → commit at intervals → run\n  `review-and-fix` over its own diff (use `pirfl` for hard\u002Fmulti-step items) → write `reports\u002F\u003Cg>.md` ending\n  `=== DONE ===`. The fixer does **not** merge.\n- Validate each wave's reports (DONE marker) before the next wave.\n\n### P6 — Gate + merge (sequential per group)\n- Merges must be serialized (one branch into master at a time). For each group with a DONE report:\n  1. Run an external review: invoke `ccc-review-cx` over the group's diff + `groups\u002F\u003Cg>.plan.md` + relevant\n     `backlog.md` rows. (If `ccc-review-cx` is unavailable, fall back to a second `review-and-fix` pass with the\n     reviewer model `openai-codex:gpt-5.5`; if no external reviewer at all, a self `review-and-fix` pass.)\n  2. **PASS** → merge `maint-\u003Crun-id>-\u003Cg>` into master, then remove the worktree (first inspect\n     `git status --porcelain --ignored` and rescue any untracked\u002Fgitignored scratch — never blind `rm -rf`\u002F`--force`).\n  3. **FAIL** → dispatch a follow-up fixer to address the review findings in the same worktree, then re-gate.\n- Append the gate result to `reports\u002F\u003Cg>.md` and mark the group merged in the ledger.\n\n### P7 — Final pass\n- Run `review-and-fix` over the whole merged diff. Run a `ccc-review-cx` review of `backlog.md` for **completeness**\n  (was every legitimate item actually fixed?). Fix gaps via a fixer agent; re-gate.\n- When the review PASSes, run `\u002Fcarm` to commit-all + rebase master. Write a final summary to `LOG.md` and set\n  `phase: done`.\n\n## Cross-cutting rules\n\n- **Resumability:** on every invoke, read the ledger and resume at the first incomplete phase. Claimed\u002Fmerged groups\n  are skipped (claims dir + ledger). Persist artifacts to disk *as each unit finishes*, not only at the end.\n- **Validate, never silently truncate:** missing output \u002F absent `=== DONE ===` ⇒ re-dispatch. If you intentionally\n  drop or defer anything (a finding, a group, a file), record it in `LOG.md` — no silent gaps.\n- **Subagents must NEVER use `attn`.** Put this line in every dispatch prompt. Agents report by writing their file.\n- **Dune\u002Ftests:** never two `dune` commands at once in the same checkout; run only the focused suites a change\n  touches; never `make test-all` in a subagent; if you must run broad tests, capture output to a temp file.\n- **Dependency floor:** the orchestration needs only bash + subagent dispatch. `review-and-fix`, `pirfl`,\n  `ccc-review-cx`, `carm` are invoked by agents and **degrade gracefully** if absent (fall back to self-review).\n\n## Agent prompts\n\nVerbatim, reusable prompt templates (Finder \u002F Triage \u002F Fixer \u002F Gate, plus the boilerplate every dispatch must carry)\nlive in [`references\u002Fprompt-pack.md`](references\u002Fprompt-pack.md). Design rationale and constraints are in\n[`references\u002Fmethodology.md`](references\u002Fmethodology.md).\n\nEvery dispatch prompt MUST include: the output file path + `=== DONE ===` contract; \"report by writing your file, not\nby messaging\"; \"**never use `attn`**\"; and the relevant dune\u002Ftest rules above.\n\n## Done checklist\n\n- [ ] `.maint\u002F` + `.worktrees\u002F` gitignored; ledger reaches `phase: done`.\n- [ ] Every target file covered by exactly one finder bucket (P1) and every finder validated (DONE).\n- [ ] `backlog.md` deduped and grouped; every legitimate item lands in a group.\n- [ ] Each group fixed in its own worktree, focused tests green, self-reviewed, gated (external review PASS), merged,\n      worktree removed.\n- [ ] Final `review-and-fix` + completeness review PASS; `\u002Fcarm` run; summary in `LOG.md`.\n",{},"master","7c87c0cf8d869c392efbe4b2f0065c85657bf969",".claude\u002Fskills\u002Fclawq-ultra-maintenance","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq","https:\u002F\u002Fclawq.org\u002F","https:\u002F\u002Fgithub.com\u002Fclankercode","xk.io","2026-11-12 11:53:34",{"source":7,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add clankercode\u002Fclawq --skill clawq-ultra-maintenance",{"tool":39,"command":40},"bunx","bunx skills add clankercode\u002Fclawq --skill clawq-ultra-maintenance",{"tool":42,"command":43},"pnpm","pnpm dlx skills add clankercode\u002Fclawq --skill clawq-ultra-maintenance","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002F.claude\u002Fskills\u002Fclawq-ultra-maintenance\u002FSKILL.md",true,[47,53],{"path":48,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":51,"sourceHtmlUrl":52},"references\u002Fmethodology.md","REFERENCE","参考文档",5917,"https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002F.claude\u002Fskills\u002Fclawq-ultra-maintenance\u002Freferences\u002Fmethodology.md",{"path":54,"resourceType":49,"resourceTypeDesc":50,"sizeBytes":55,"sourceHtmlUrl":56},"references\u002Fprompt-pack.md",4487,"https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002F.claude\u002Fskills\u002Fclawq-ultra-maintenance\u002Freferences\u002Fprompt-pack.md",[58,90,119,129,150,161],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":69,"publishedAt":20},"01M6CZGG2GJZ7PGVTZSG8NDB2R","run-task","Runs a backlog item (bug, feature, task, epic, milestone, phase) end-to-end by gathering context, constructing a startup prompt, spawning a worktree-backed subagent, and managing the claim\u002Fdone\u002Freview cycle. Use when the user wants to execute a backlog target like B123, F456, P2.M4.E3, etc.","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002Fdocs\u002Fbuiltin-skills\u002Frun-task\u002FSKILL.md",[64],{"categoryNo":65,"categoryName":66,"taxonomyType":67,"taxonomyTypeDesc":68},"01JBSEEDCAT000000000000023","Skill与智能体开发","SCENARIO","使用场景",[70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89],"run task","run","task","backlog","feature","milestone","review cycle","review","done","worktree","startup","subagent","extra","prompt","user","execute","use","bug","claim","cycle",{"skillNo":91,"name":92,"nameZh":6,"description":93,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":94,"categories":95,"keywords":99,"publishedAt":20},"01M6CZGD2YWGXTV25FM004F7ZM","pipeline-designer","Design and create structured output pipelines. Gathers requirements, designs step sequences with output schemas, generates valid YAML pipeline definitions, and validates them.","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002Fdocs\u002Fbuiltin-skills\u002Fpipeline-designer\u002FSKILL.md",[96],{"categoryNo":97,"categoryName":98,"taxonomyType":67,"taxonomyTypeDesc":68},"01JBSEEDCAT000000000000015","视觉设计",[100,101,102,103,104,105,106,107,84,108,109,110,111,112,113,114,115,116,117,118],"pipeline","designer","pipeline designer","read","message","what the","do","pipelines","write","requirements","question","ask","shell","valid","structured","file","create","design","yaml",{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":20},"01M6CZGAY23A6HQS1G7JK7WQGM","idea","Log a planning idea to the backlog. Use when the user has a feature idea, enhancement suggestion, or brainstorm they want tracked.","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002Fdocs\u002Fbuiltin-skills\u002Fidea\u002FSKILL.md",[],[121,126,74,73,127,86,84,128],"brainstorm","log","planning",{"skillNo":130,"name":74,"nameZh":6,"description":131,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":132,"categories":133,"keywords":135,"publishedAt":20},"01M6CZG7S5TSVHB6S8YA3JMGFJ","Automate the creation of a new feature for clawq. Gathers requirements, discovers existing capabilities, decomposes into implementation layers, gates risky work behind approval, and delegates implementation. Prefers skills\u002Frigs\u002Fcron\u002Ftemplates over OCaml code changes.","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002Fdocs\u002Fbuiltin-skills\u002Ffeature\u002FSKILL.md",[134],{"categoryNo":65,"categoryName":66,"taxonomyType":67,"taxonomyTypeDesc":68},[74,136,137,138,139,140,141,142,143,144,145,18,146,147,109,148,149],"code","implementation","approval","new feature","work","automate","cron","templates","capabilities","ocaml","creation","gates","natural language","language",{"skillNo":151,"name":152,"nameZh":6,"description":153,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":154,"categories":155,"keywords":156,"publishedAt":20},"01M6CZG4STQ8R15F6HFT5BE98J","telegram-formatting","Telegram HTML and MarkdownV2 formatting reference for clawq message code","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002F.claude\u002Fskills\u002Ftelegram-formatting\u002FSKILL.md",[],[157,158,18,104,159,160,136],"telegram formatting","telegram","reference","html",{"skillNo":162,"name":163,"nameZh":6,"description":164,"descriptionZh":6,"repositoryFullName":7,"ownerLogin":8,"ownerAvatarUrl":9,"ownerAccountNo":10,"ownerDisplayName":6,"repositoryNo":11,"starsCount":12,"licenseSpdxId":13,"licenseFileUrl":6,"sourceHtmlUrl":165,"categories":166,"keywords":167,"publishedAt":20},"01M6CZG2ZASD966C153XZY3BQ1","pag-review-via-team","review work using teams + review-and-fix","https:\u002F\u002Fgithub.com\u002Fclankercode\u002Fclawq\u002Fblob\u002Fmaster\u002F.claude\u002Fskills\u002Fpag-review-via-team\u002FSKILL.md",[],[77,168,169,170,140,171],"pag review via team","team","fix","teams"]