[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M6GNGX7GRP4RW8V2EQ0DGT8B":3,"skill-same:01M6GNGX7GRP4RW8V2EQ0DGT8B":118},{"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":6,"sourceHtmlUrl":16,"categories":17,"keywords":23,"publishedAt":33,"linkOnly":34,"markdownContent":35,"markdownBody":36,"frontMatter":37,"refName":40,"commitSha":41,"skillDirectory":42,"forksCount":43,"repositoryHtmlUrl":44,"repositoryHomepageUrl":6,"ownerHtmlUrl":45,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":46,"install":47,"resources":59},"01M6GNGX7GRP4RW8V2EQ0DGT8B","pdca",null,"Use when running a PDCA cycle across research, production, review, and refinement","unclejobs-ai\u002Fsecond-claude-code","unclejobs-ai","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F150198827?v=4","01M6GNFYGQ7354R1N04YF1B269","Unclejobs","01M6GNFYNCQEKYTVX572HB0F26",41,"MIT","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002FSKILL.md",[18],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000021","资料检索与研究","SCENARIO","使用场景",[5,24,25,26,27,28,29,30,31,32],"review","research","Production","use","cycle","across","false","refinement","true","2026-11-10 10:12:30",false,"---\nname: pdca\ndescription: \"Use when running a PDCA cycle across research, production, review, and refinement\"\neffort: high\ndisable-model-invocation: true\nuser-invocable: false\n---\n\n## Iron Law\n\n> **When a PDCA run is selected, honor its phase gates and transition order.**\n\n## Red Flags\n\n- \"I can write tests later\" → STOP. Write them now.\n- \"This change is too small to review\" → STOP. Small bugs become P0 incidents.\n- \"I don't need to check previous cycle insights\" → STOP. You will repeat the same mistake.\n- \"This is good enough\" → STOP. Check the checklist.\n- \"No time to follow every step\" → STOP. Skipped steps cost 3x more later.\n\n# PDCA — Knowledge Work Cycle Orchestrator\n\nMeta-skill that runs the Plan → Do → Check → Act cycle.\nEach phase gates into the next. No gate skipping.\n\n## When to Use\n\n- User wants end-to-end knowledge work (research + write + review + improve)\n- User mentions a PDCA phase explicitly (\"plan this\", \"check this\")\n- Complex work requiring multiple skills in sequence\n- User says \"알아보고 써줘\", \"research and write\", or similar multi-phase requests\n\n## PDCA Is an Optional Orchestration Path\n\nUse PDCA for end-to-end or explicitly phased work. Individual skills and commands such as\n`\u002Fscc:research`, `\u002Fscc:write`, `\u002Fscc:review`, and `\u002Fscc:refine` also run directly; they are\nnot required to be wrapped in PDCA.\n\n```\nPDCA Cycle (when selected)\n  ├─ Plan  → calls \u002Fscc:research (or domain research) + \u002Fscc:analyze\n  ├─ Do    → calls the appropriate sub-skill based on output format\n  │           └─ Supported format?     → \u002Fscc:write --format \u003Cformat>\n  ├─ Check → calls \u002Fscc:review (always, regardless of which sub-skill ran in Do)\n  └─ Act   → calls \u002Fscc:refine or routes back to Plan\u002FDo via Action Router\n```\n\nThe key principle: PDCA owns transitions only for a PDCA run. It chooses a supported format\nand delegates production, then validates the resulting phase output.\n\n### Code Engineering Lane\n\nWhen the task is code work (`domain=code`, repository edits, tests, debugging, CI, refactoring, or PR prep), PDCA still owns the Plan -> Do -> Check -> Act cycle, but it must load `references\u002Fcode-engineering-lane.md` before planning. This lane absorbs the useful code-work discipline from external playbooks while preserving Second Claude's existing PDCA architecture.\n\nThe lane adds code-specific gates:\n\n- Plan must produce executable acceptance criteria before medium or high-risk edits.\n- Do should use branch\u002Fworktree isolation for non-trivial implementation and keep a stage report for long work.\n- Check must separate worker output from validator\u002Freviewer proof.\n- Act must clean, simplify, and record handoff state before completion.\n\nUse the lane as a specialization layer, not as a replacement for `review`, `workflow`, or `loop`.\n\n### Do-Phase Sub-Skill Selection\n\nWhen PDCA enters the Do phase, it picks the most specialized sub-skill that matches the artifact format:\n\n| Output Format | Sub-Skill PDCA Dispatches | What That Sub-Skill Owns Internally |\n|---------------|--------------------------|-------------------------------------|\n| `newsletter`, `article`, `shorts`, `report`, `social`, `card-news` | `\u002Fscc:write --format \u003Cformat>` | Format-specific production contract |\n| Other specialized format supplied by an installed plugin | External plugin capability (optional) | Only when the caller explicitly invokes it |\n| Generic content | `\u002Fscc:write` | Pure execution from Plan artifacts |\n\n**Key trigger keywords for sub-skill selection** (PDCA scans the user's prompt and the Plan output's `dod` field):\n\n- \"뉴스레터\", \"newsletter\" → `\u002Fscc:write --format newsletter` in Do\n- \"아티클\", \"article\" → `\u002Fscc:write --format article` in Do\n- \"쇼츠\", \"shorts\", \"릴스\", \"Reels\" → `\u002Fscc:write --format shorts` in Do\n- \"보고서\", \"report\" → `\u002Fscc:write --format report` in Do\n- \"소셜\", \"social\" → `\u002Fscc:write --format social` in Do\n- \"카드뉴스\", \"card news\", \"캐러셀\" → `\u002Fscc:write --format card-news` in Do\n- Anything else → `\u002Fscc:write` in Do\n\n### Why PDCA Wraps Sub-Skills (Doesn't Hand Off)\n\nThe Plan and Check phases add value that no sub-skill provides on its own:\n\n- **Plan adds upstream rigor**: Question Protocol, source minimums, framework analysis, DoD definition. Without Plan, sub-skills work on whatever scope the user happened to specify, which is often incomplete.\n- **Check adds downstream rigor**: Multi-reviewer parallel validation, fact-checking, structural analysis, cross-model review. Without Check, sub-skills self-approve their own output, which misses systematic blind spots.\n- **Act adds correction loop**: Action Router classifies failures by root cause (research gap vs format gap vs polish gap) and routes back to the right phase. Without Act, the user has to manually decide what to fix.\n\nA sub-skill running on its own = single-shot generation with the sub-skill's internal checks only.\nPDCA wrapping a sub-skill = upstream Plan rigor + sub-skill's internal pipeline + downstream Check + Act loop. Strictly more validation.\n\n### When an Optional External Capability Has Its Own Phases\n\nAn installed external plugin may expose a specialized capability with its own phases. If a\ncaller explicitly invokes it, those phases are outside this plugin's built-in routing; PDCA\ncan validate the returned artifact when it is used as Do input.\n\nFor the full sub-skill dispatch protocol (input contract, output contract, failure handling, integration points), see `references\u002Fdomain-pipeline-integration.md`.\n\nThis means a full PDCA run on a newsletter looks like:\n\n```\nPDCA Plan  →  \u002Fscc:research + \u002Fscc:analyze (gather sources, build framework)\nPDCA Do    →  \u002Fscc:write --format newsletter (research\u002Freview can be skipped when\n              Plan and Check are owned by this PDCA run)\nPDCA Check →  \u002Fscc:review (parallel reviewers; this is the independent Check phase)\nPDCA Act   →  Action Router classifies any Check findings and routes to Plan,\n              Do, or Refine\n```\n\nPDCA's Check is independent of any review a caller or optional external capability may have\nperformed. It supplies the cycle's configured consensus verdict.\n\n### Past Failure That Motivated This Architecture\n\nThe lesson: PDCA adds value when the user needs phase gates and routed correction. It is not a\nrequirement for every direct skill invocation, and this plugin does not assume unavailable\nspecialized commands.\n\n## Phase Detection\n\n| Signal | Phase | Skills Chained |\n|--------|-------|---------------|\n| \"조사해\", \"알아봐\", plan, research, explore | **Plan** (Gather) | research → analyze |\n| \"써\", \"만들어\", \"분석해\", create, write, build | **Do** (Produce) | write (pure execution) |\n| \"검토해\", \"리뷰해\", review, check, verify | **Check** (Verify) | review |\n| \"고쳐\", \"개선해\", improve, fix, refine, iterate | **Act** (Refine) | action router → refine |\n| \"알아보고 써줘\", \"end-to-end\", \"full report\" | **Full PDCA** | All phases in sequence |\n\nWhen a single phase is detected, run that phase and pause at its gate.\nWhen full PDCA is detected, run all phases with gates between each.\n\n## Architecture\n\n```\nUser Prompt → normal Claude Code skill\u002Fcommand selection\n\nPDCA Orchestrator\n│\n├── PLAN (research → analyze)\n│   ├── Question Protocol (max 3 Qs → unanswered = save assumptions)\n│   ├── Eevee (researcher): data collection\n│   └── Alakazam (analyst) + Mewtwo (strategist): structured analysis\n│       └── Gate: Brief + Analysis exist? Runtime source-count gate (≥5)\n│\n├── DO (pure execution)\n│   └── Smeargle (writer): \u002Fscc:write --skip-research --skip-review\n│       └── Gate: Artifact complete? Format OK?\n│\n├── CHECK (unchanged)\n│   └── Xatu + Absol + Porygon + Jigglypuff + Unown: \u002Fscc:review\n│       └── Gate: Verdict routing\n│\n└── ACT (action router → refine)\n    ├── Action Router (root cause classification)\n    │   ├── SOURCE\u002FASSUMPTION\u002FFRAMEWORK → PLAN\n    │   ├── COMPLETENESS\u002FFORMAT → DO\n    │   └── EXECUTION_QUALITY → REFINE\n    └── Ditto (editor): \u002Fscc:refine\n        └── Gate: Target met? → EXIT\n```\n\n## Phase Output Schemas\n\nEach phase outputs a validated schema before the gate is passed. See `references\u002Fphase-schemas.md` for full field rules and validation failure actions. Missing required fields are gate failures, not warnings.\n\n## Phase Gates\n\nLoad the relevant checklist from `references\u002F` at each transition. Gates are mandatory — do NOT skip.\n\n| Gate | Reference | Permission | Key Failure Condition |\n|------|-----------|------------|----------------------|\n| Plan → Do | `references\u002Fplan-phase.md` | `plan` (read-only) | runtime `sources_count \u003C 5`, Plan Mode not approved, any required field missing |\n| Do → Check | `references\u002Fdo-phase.md` | `acceptEdits` | `plan_findings_integrated: false`, `sections_complete: false`, artifact absent |\n| Check → Act | `references\u002Fcheck-phase.md` | `plan` (read-only) | Fewer than 2 reviewers, verdict not a standard value |\n| Act → Exit\u002FCycle | `references\u002Fact-phase.md` | `acceptEdits` | `decision` invalid, `root_cause_category` empty |\n\n**Permission pattern**: read-only (`plan`) → write (`acceptEdits`) → read-only → write. Switch at every phase boundary.\n\n**Check → Act routing**: `APPROVED` → EXIT | `MINOR FIXES` → Act light touch | `NEEDS IMPROVEMENT` → Act full refine | `MUST FIX` → Act critical-first. Preset reviewer counts and optional quality checks remain defined by `\u002Fscc:review`; not every narrative contract is enforced by the state MCP runtime.\n\n**Cycle limit**: If `cycle_count >= max_cycles`, do NOT start another cycle. Notify: \"max_cycles에 도달했습니다 — 현재 결과물로 종료합니다.\" Exit with best artifact.\n\n**Worktree**: Do artifacts written in `worktree-pdca-do`. If Check returns MUST FIX, orchestrator discards the worktree for a clean restart.\n\n## Workflow (Full PDCA)\n\n1. **Plan**: Question Protocol → research (Eevee) → analyze (Alakazam + Mewtwo) → **EnterPlanMode** (brief user) → **ExitPlanMode** (get approval). On rejection: re-run with feedback.\n2. **Gate**: brief + analysis quality + Plan Mode approval.\n3. **Do**: write (Smeargle) `--skip-research --skip-review` using Plan artifacts.\n4. **Gate**: artifact completeness.\n5. **Check**: review (Xatu, Absol, Porygon, Jigglypuff, Unown) with appropriate preset.\n6. **Gate**: verdict → Act or Exit.\n7. **Act**: Action Router → Plan \u002F Do \u002F Loop.\n8. **Gate**: target met → Exit or present options.\n\n## Options\n\n| Flag | Values | Default |\n|------|--------|---------|\n| `--phase` | `plan\\|do\\|check\\|act\\|full` | auto-detect |\n| `--depth` | `shallow\\|medium\\|deep` | `medium` |\n| `--target` | verdict or score | `APPROVED` |\n| `--max` | max Act iterations | `3` |\n| `--max-cycles` | max full PDCA re-cycles (Act→Plan→Do→Check) | `3` |\n| `--no-questions` | skip Question Protocol | `false` |\n\nNote: `--constraints` is a write-skill flag passed through by the orchestrator (see do-phase.md, act-phase.md).\n\n## Stuck Detection\n\nRun at every phase transition. See `references\u002Fstuck-detection.md` for full signal definitions, root causes, and remediation steps.\n\n| Pattern | Trigger | Action |\n|---------|---------|--------|\n| **Plan Churn** | `cycle_count >= 3` AND `\"do\"` not in `completed[]` | Force Do with current plan; append uncertainty note |\n| **Check Avoidance** | `\"do\"` in `completed[]` AND `artifacts.check_report` is null before entering Act | Block Act; inject DoD checklist; require review dispatch |\n| **Scope Creep** | Do artifact scope diverges significantly from Plan scope at Do→Check gate | Alert user with planned-vs-actual comparison; wait for choice |\n\nIf both stuck patterns and `max_cycles` trigger simultaneously, report both and exit with the best artifact.\n\n## State\n\nState persisted in `.data\u002Fstate\u002Fpdca-active.json`. See `references\u002Fstate-schema.md` for full schema, field definitions, and single-active-run constraint.\n\nMCP tools (when `pdca-state` server is available): `pdca_get_state`, `pdca_start_run`, `pdca_transition`, `pdca_check_gate`, `pdca_update_stuck_flags`, `pdca_end_run`.\n\n## Output\n\nAt each gate: phase summary + gate verdict (pass\u002Ffail + reason) + next phase recommendation.\n\nAt cycle end: full artifact chain, Action Router decisions, verdict progression, total phases + token cost estimate.\n\n## Notifications\n\nWhen Channels are configured (via `.data\u002Fchannels.json` or `TELEGRAM_CHAT_ID` env), PDCA emits\nnotifications at phase transitions and review verdicts. Opt-in via `--notify` flag or by\ncreating `.data\u002Fchannels.json` from the `.data\u002Fchannels.json.example` template.\n\nSee `references\u002Fchannels-integration.md` for event formats, configuration fields, and\nthe notification payload pattern used by `hooks\u002Fsession-end.mjs`.\n\n## Subagents (Conceptual Roles, Not Direct Dispatch Targets)\n\n**Important**: The Pokemon names below are **conceptual role labels**, not Agent tool `subagent_type` values. PDCA does NOT dispatch them directly via the Agent tool. Instead, the chained skills (`\u002Fscc:research`, `\u002Fscc:write`, `\u002Fscc:review`, `\u002Fscc:refine`) handle these roles internally — they map each role to the appropriate concrete subagent or model.\n\nIf you (the orchestrator) try to call `Agent(subagent_type: \"eevee\")`, it will fail. Always go through the chained skill, never bypass to a Pokemon role.\n\n```yaml\n# Plan dispatches two skills, so its roles split across them.\n# Roles handled inside \u002Fscc:research\nresearcher: { model: sonnet, role: eevee, purpose: \"collect sources with citations, minimum 5 distinct sources\" }\nanalyst: { model: sonnet, role: alakazam, purpose: \"list gaps, flag data conflicts, verify coverage\" }\n\n# Roles handled inside \u002Fscc:analyze — alakazam appears in both, pre-processing there rather than gap-listing\nanalyst: { model: sonnet, role: alakazam, purpose: \"structure raw research into patterns before a framework is applied\" }\nstrategist: { model: sonnet, role: mewtwo, purpose: \"apply the framework, cite evidence, challenge assumptions\" }\n\n# Role handled inside \u002Fscc:write\nwriter: { model: opus, role: smeargle, purpose: \"produce artifact from plan, skip-research skip-review\" }\n\n# Roles handled inside \u002Fscc:review (parallel dispatch)\ndeep-reviewer: { model: opus, role: xatu, purpose: \"thorough quality review with structured critic output\" }\ndevil-advocate: { model: sonnet, role: absol, purpose: \"attack weakest points, find logical gaps\" }\nfact-checker: { model: sonnet, role: porygon, purpose: \"verify claims against sources, flag unsupported statements\" }\ntone-guardian: { model: sonnet, role: jigglypuff, purpose: \"check voice and audience fit against .data\u002Fsoul\u002FSOUL.md Tone Rules and Anti-Patterns\" }\nstructure-analyst: { model: sonnet, role: unown, purpose: \"check organization, flow, format compliance\" }\n\n# Role handled inside \u002Fscc:refine\neditor: { model: opus, role: ditto, purpose: \"apply top 3 fixes per iteration, verify improvement\" }\n\n# Orchestration meta-role (the PDCA orchestrator itself)\npipeline-orchestrator: { model: sonnet, role: arceus, purpose: \"enforce gates, manage phase transitions, never skip phases, prefer domain hand-off over self-processing\" }\n```\n\n**How PDCA actually executes**: The orchestrator calls `\u002Fscc:research`, `\u002Fscc:write`, `\u002Fscc:review`, `\u002Fscc:refine` as Skill invocations. Each of those skills internally dispatches the right subagents (`general-purpose`, `code-reviewer`, etc.) using the Agent tool. PDCA never bypasses this layer.\n\n**Tier overrides are upward-only.** `--tier fable` raises the two spine roles — `writer` (smeargle) and `deep-reviewer` (xatu) — for high-stakes runs. Nothing lowers them: not a flag, and not an automatic fallback when a dispatch fails. If those dispatches are exhausted, escalate to the user instead of retrying at a cheaper tier, because every later phase inherits what the writer and the whole-document reviewer missed. To make a run cheaper, cut cycles or reviewers. See `references\u002Fdesign-principles.md` §10.\n\n## Gotchas\n\n- Do NOT skip gates. They prevent garbage-in-garbage-out.\n- Do NOT run Do without Plan unless user explicitly has source material ready.\n- Do NOT declare complete without Check phase verdict.\n- Do NOT route all Act findings to Loop — use the Action Router to classify root causes.\n- Full PDCA with deep research = significant token cost. Warn user at start.\n\nSee `gotchas.md` for extended gotchas (stuck detection, worktree lifecycle, permission transitions, session resume, automation mode).\n\n## MMBridge Session Handoff\n\nWhen mmbridge is detected (see `references\u002Fmmbridge-integration.md`) and the PDCA cycle exits with `APPROVED`, generate a handoff artifact.\n\n### When to Trigger\n\n- PDCA cycle reaches EXIT (APPROVED verdict from Check phase)\n- mmbridge is installed\n- Do NOT trigger on partial cycles or early exits\n\n### Command\n\n```bash\nmmbridge handoff --write .captures\u002Fpdca-handoff-${RUN_ID}.md\n```\n\n### Output\n\nThe handoff artifact summarizes:\n- What was researched (Plan phase sources and findings)\n- What was produced (Do phase artifact path and type)\n- What reviewers found (Check phase verdict and key findings)\n- What was refined (Act phase changes, if any)\n\nPresent the handoff path to the user:\n> \"PDCA cycle complete. Handoff artifact saved to `.captures\u002Fpdca-handoff-${RUN_ID}.md`\"\n\n### Usage\n\nThe handoff artifact serves two purposes:\n1. **New session briefing**: Load into a fresh Claude session for continuity\n2. **Team communication**: Share with colleagues as a work summary\n\nThis is an **optional convenience** — the PDCA cycle is complete regardless of whether the handoff succeeds.\n\n## MMBridge Embrace (Full-Cycle Acceleration)\n\nWhen mmbridge is detected, the PDCA orchestrator can optionally delegate the full cycle to `mmbridge embrace` for parallel multi-model acceleration.\n\n### When to Use\n\n- `--depth deep` with full PDCA: embrace provides multi-model research and debate in parallel with internal agents\n- Large-scope tasks where multi-model perspectives add value\n- NOT for `--depth shallow` or single-phase runs\n\n### Dispatch\n\nAt the start of a full PDCA cycle, if mmbridge is detected and depth is `deep`:\n\n```bash\nmmbridge embrace \"\u003Ctask summary>\" --json --non-interactive > \u002Ftmp\u002Fmmbridge-embrace-${RUN_ID}.json\n```\n\n- `--non-interactive`: auto-proceed through checkpoints (PDCA orchestrator manages gates, not embrace)\n- `--skip-phases`: use when specific PDCA phases should not have embrace acceleration (e.g., `--skip-phases security` if not a code project)\n\n### Merge Points\n\nEmbrace results are merged at multiple PDCA phases:\n- **Plan**: embrace research findings → supplemental sources for Eevee\n- **Check**: embrace review findings → additional voter in consensus gate\n- **Act**: embrace security findings → additional input for action router\n\n### Cost Note\n\n`mmbridge embrace` runs multiple models across multiple phases. Only activate at `deep` depth. Warn user about estimated cost at PDCA start.\n","## Iron Law\n\n> **When a PDCA run is selected, honor its phase gates and transition order.**\n\n## Red Flags\n\n- \"I can write tests later\" → STOP. Write them now.\n- \"This change is too small to review\" → STOP. Small bugs become P0 incidents.\n- \"I don't need to check previous cycle insights\" → STOP. You will repeat the same mistake.\n- \"This is good enough\" → STOP. Check the checklist.\n- \"No time to follow every step\" → STOP. Skipped steps cost 3x more later.\n\n# PDCA — Knowledge Work Cycle Orchestrator\n\nMeta-skill that runs the Plan → Do → Check → Act cycle.\nEach phase gates into the next. No gate skipping.\n\n## When to Use\n\n- User wants end-to-end knowledge work (research + write + review + improve)\n- User mentions a PDCA phase explicitly (\"plan this\", \"check this\")\n- Complex work requiring multiple skills in sequence\n- User says \"알아보고 써줘\", \"research and write\", or similar multi-phase requests\n\n## PDCA Is an Optional Orchestration Path\n\nUse PDCA for end-to-end or explicitly phased work. Individual skills and commands such as\n`\u002Fscc:research`, `\u002Fscc:write`, `\u002Fscc:review`, and `\u002Fscc:refine` also run directly; they are\nnot required to be wrapped in PDCA.\n\n```\nPDCA Cycle (when selected)\n  ├─ Plan  → calls \u002Fscc:research (or domain research) + \u002Fscc:analyze\n  ├─ Do    → calls the appropriate sub-skill based on output format\n  │           └─ Supported format?     → \u002Fscc:write --format \u003Cformat>\n  ├─ Check → calls \u002Fscc:review (always, regardless of which sub-skill ran in Do)\n  └─ Act   → calls \u002Fscc:refine or routes back to Plan\u002FDo via Action Router\n```\n\nThe key principle: PDCA owns transitions only for a PDCA run. It chooses a supported format\nand delegates production, then validates the resulting phase output.\n\n### Code Engineering Lane\n\nWhen the task is code work (`domain=code`, repository edits, tests, debugging, CI, refactoring, or PR prep), PDCA still owns the Plan -> Do -> Check -> Act cycle, but it must load `references\u002Fcode-engineering-lane.md` before planning. This lane absorbs the useful code-work discipline from external playbooks while preserving Second Claude's existing PDCA architecture.\n\nThe lane adds code-specific gates:\n\n- Plan must produce executable acceptance criteria before medium or high-risk edits.\n- Do should use branch\u002Fworktree isolation for non-trivial implementation and keep a stage report for long work.\n- Check must separate worker output from validator\u002Freviewer proof.\n- Act must clean, simplify, and record handoff state before completion.\n\nUse the lane as a specialization layer, not as a replacement for `review`, `workflow`, or `loop`.\n\n### Do-Phase Sub-Skill Selection\n\nWhen PDCA enters the Do phase, it picks the most specialized sub-skill that matches the artifact format:\n\n| Output Format | Sub-Skill PDCA Dispatches | What That Sub-Skill Owns Internally |\n|---------------|--------------------------|-------------------------------------|\n| `newsletter`, `article`, `shorts`, `report`, `social`, `card-news` | `\u002Fscc:write --format \u003Cformat>` | Format-specific production contract |\n| Other specialized format supplied by an installed plugin | External plugin capability (optional) | Only when the caller explicitly invokes it |\n| Generic content | `\u002Fscc:write` | Pure execution from Plan artifacts |\n\n**Key trigger keywords for sub-skill selection** (PDCA scans the user's prompt and the Plan output's `dod` field):\n\n- \"뉴스레터\", \"newsletter\" → `\u002Fscc:write --format newsletter` in Do\n- \"아티클\", \"article\" → `\u002Fscc:write --format article` in Do\n- \"쇼츠\", \"shorts\", \"릴스\", \"Reels\" → `\u002Fscc:write --format shorts` in Do\n- \"보고서\", \"report\" → `\u002Fscc:write --format report` in Do\n- \"소셜\", \"social\" → `\u002Fscc:write --format social` in Do\n- \"카드뉴스\", \"card news\", \"캐러셀\" → `\u002Fscc:write --format card-news` in Do\n- Anything else → `\u002Fscc:write` in Do\n\n### Why PDCA Wraps Sub-Skills (Doesn't Hand Off)\n\nThe Plan and Check phases add value that no sub-skill provides on its own:\n\n- **Plan adds upstream rigor**: Question Protocol, source minimums, framework analysis, DoD definition. Without Plan, sub-skills work on whatever scope the user happened to specify, which is often incomplete.\n- **Check adds downstream rigor**: Multi-reviewer parallel validation, fact-checking, structural analysis, cross-model review. Without Check, sub-skills self-approve their own output, which misses systematic blind spots.\n- **Act adds correction loop**: Action Router classifies failures by root cause (research gap vs format gap vs polish gap) and routes back to the right phase. Without Act, the user has to manually decide what to fix.\n\nA sub-skill running on its own = single-shot generation with the sub-skill's internal checks only.\nPDCA wrapping a sub-skill = upstream Plan rigor + sub-skill's internal pipeline + downstream Check + Act loop. Strictly more validation.\n\n### When an Optional External Capability Has Its Own Phases\n\nAn installed external plugin may expose a specialized capability with its own phases. If a\ncaller explicitly invokes it, those phases are outside this plugin's built-in routing; PDCA\ncan validate the returned artifact when it is used as Do input.\n\nFor the full sub-skill dispatch protocol (input contract, output contract, failure handling, integration points), see `references\u002Fdomain-pipeline-integration.md`.\n\nThis means a full PDCA run on a newsletter looks like:\n\n```\nPDCA Plan  →  \u002Fscc:research + \u002Fscc:analyze (gather sources, build framework)\nPDCA Do    →  \u002Fscc:write --format newsletter (research\u002Freview can be skipped when\n              Plan and Check are owned by this PDCA run)\nPDCA Check →  \u002Fscc:review (parallel reviewers; this is the independent Check phase)\nPDCA Act   →  Action Router classifies any Check findings and routes to Plan,\n              Do, or Refine\n```\n\nPDCA's Check is independent of any review a caller or optional external capability may have\nperformed. It supplies the cycle's configured consensus verdict.\n\n### Past Failure That Motivated This Architecture\n\nThe lesson: PDCA adds value when the user needs phase gates and routed correction. It is not a\nrequirement for every direct skill invocation, and this plugin does not assume unavailable\nspecialized commands.\n\n## Phase Detection\n\n| Signal | Phase | Skills Chained |\n|--------|-------|---------------|\n| \"조사해\", \"알아봐\", plan, research, explore | **Plan** (Gather) | research → analyze |\n| \"써\", \"만들어\", \"분석해\", create, write, build | **Do** (Produce) | write (pure execution) |\n| \"검토해\", \"리뷰해\", review, check, verify | **Check** (Verify) | review |\n| \"고쳐\", \"개선해\", improve, fix, refine, iterate | **Act** (Refine) | action router → refine |\n| \"알아보고 써줘\", \"end-to-end\", \"full report\" | **Full PDCA** | All phases in sequence |\n\nWhen a single phase is detected, run that phase and pause at its gate.\nWhen full PDCA is detected, run all phases with gates between each.\n\n## Architecture\n\n```\nUser Prompt → normal Claude Code skill\u002Fcommand selection\n\nPDCA Orchestrator\n│\n├── PLAN (research → analyze)\n│   ├── Question Protocol (max 3 Qs → unanswered = save assumptions)\n│   ├── Eevee (researcher): data collection\n│   └── Alakazam (analyst) + Mewtwo (strategist): structured analysis\n│       └── Gate: Brief + Analysis exist? Runtime source-count gate (≥5)\n│\n├── DO (pure execution)\n│   └── Smeargle (writer): \u002Fscc:write --skip-research --skip-review\n│       └── Gate: Artifact complete? Format OK?\n│\n├── CHECK (unchanged)\n│   └── Xatu + Absol + Porygon + Jigglypuff + Unown: \u002Fscc:review\n│       └── Gate: Verdict routing\n│\n└── ACT (action router → refine)\n    ├── Action Router (root cause classification)\n    │   ├── SOURCE\u002FASSUMPTION\u002FFRAMEWORK → PLAN\n    │   ├── COMPLETENESS\u002FFORMAT → DO\n    │   └── EXECUTION_QUALITY → REFINE\n    └── Ditto (editor): \u002Fscc:refine\n        └── Gate: Target met? → EXIT\n```\n\n## Phase Output Schemas\n\nEach phase outputs a validated schema before the gate is passed. See `references\u002Fphase-schemas.md` for full field rules and validation failure actions. Missing required fields are gate failures, not warnings.\n\n## Phase Gates\n\nLoad the relevant checklist from `references\u002F` at each transition. Gates are mandatory — do NOT skip.\n\n| Gate | Reference | Permission | Key Failure Condition |\n|------|-----------|------------|----------------------|\n| Plan → Do | `references\u002Fplan-phase.md` | `plan` (read-only) | runtime `sources_count \u003C 5`, Plan Mode not approved, any required field missing |\n| Do → Check | `references\u002Fdo-phase.md` | `acceptEdits` | `plan_findings_integrated: false`, `sections_complete: false`, artifact absent |\n| Check → Act | `references\u002Fcheck-phase.md` | `plan` (read-only) | Fewer than 2 reviewers, verdict not a standard value |\n| Act → Exit\u002FCycle | `references\u002Fact-phase.md` | `acceptEdits` | `decision` invalid, `root_cause_category` empty |\n\n**Permission pattern**: read-only (`plan`) → write (`acceptEdits`) → read-only → write. Switch at every phase boundary.\n\n**Check → Act routing**: `APPROVED` → EXIT | `MINOR FIXES` → Act light touch | `NEEDS IMPROVEMENT` → Act full refine | `MUST FIX` → Act critical-first. Preset reviewer counts and optional quality checks remain defined by `\u002Fscc:review`; not every narrative contract is enforced by the state MCP runtime.\n\n**Cycle limit**: If `cycle_count >= max_cycles`, do NOT start another cycle. Notify: \"max_cycles에 도달했습니다 — 현재 결과물로 종료합니다.\" Exit with best artifact.\n\n**Worktree**: Do artifacts written in `worktree-pdca-do`. If Check returns MUST FIX, orchestrator discards the worktree for a clean restart.\n\n## Workflow (Full PDCA)\n\n1. **Plan**: Question Protocol → research (Eevee) → analyze (Alakazam + Mewtwo) → **EnterPlanMode** (brief user) → **ExitPlanMode** (get approval). On rejection: re-run with feedback.\n2. **Gate**: brief + analysis quality + Plan Mode approval.\n3. **Do**: write (Smeargle) `--skip-research --skip-review` using Plan artifacts.\n4. **Gate**: artifact completeness.\n5. **Check**: review (Xatu, Absol, Porygon, Jigglypuff, Unown) with appropriate preset.\n6. **Gate**: verdict → Act or Exit.\n7. **Act**: Action Router → Plan \u002F Do \u002F Loop.\n8. **Gate**: target met → Exit or present options.\n\n## Options\n\n| Flag | Values | Default |\n|------|--------|---------|\n| `--phase` | `plan\\|do\\|check\\|act\\|full` | auto-detect |\n| `--depth` | `shallow\\|medium\\|deep` | `medium` |\n| `--target` | verdict or score | `APPROVED` |\n| `--max` | max Act iterations | `3` |\n| `--max-cycles` | max full PDCA re-cycles (Act→Plan→Do→Check) | `3` |\n| `--no-questions` | skip Question Protocol | `false` |\n\nNote: `--constraints` is a write-skill flag passed through by the orchestrator (see do-phase.md, act-phase.md).\n\n## Stuck Detection\n\nRun at every phase transition. See `references\u002Fstuck-detection.md` for full signal definitions, root causes, and remediation steps.\n\n| Pattern | Trigger | Action |\n|---------|---------|--------|\n| **Plan Churn** | `cycle_count >= 3` AND `\"do\"` not in `completed[]` | Force Do with current plan; append uncertainty note |\n| **Check Avoidance** | `\"do\"` in `completed[]` AND `artifacts.check_report` is null before entering Act | Block Act; inject DoD checklist; require review dispatch |\n| **Scope Creep** | Do artifact scope diverges significantly from Plan scope at Do→Check gate | Alert user with planned-vs-actual comparison; wait for choice |\n\nIf both stuck patterns and `max_cycles` trigger simultaneously, report both and exit with the best artifact.\n\n## State\n\nState persisted in `.data\u002Fstate\u002Fpdca-active.json`. See `references\u002Fstate-schema.md` for full schema, field definitions, and single-active-run constraint.\n\nMCP tools (when `pdca-state` server is available): `pdca_get_state`, `pdca_start_run`, `pdca_transition`, `pdca_check_gate`, `pdca_update_stuck_flags`, `pdca_end_run`.\n\n## Output\n\nAt each gate: phase summary + gate verdict (pass\u002Ffail + reason) + next phase recommendation.\n\nAt cycle end: full artifact chain, Action Router decisions, verdict progression, total phases + token cost estimate.\n\n## Notifications\n\nWhen Channels are configured (via `.data\u002Fchannels.json` or `TELEGRAM_CHAT_ID` env), PDCA emits\nnotifications at phase transitions and review verdicts. Opt-in via `--notify` flag or by\ncreating `.data\u002Fchannels.json` from the `.data\u002Fchannels.json.example` template.\n\nSee `references\u002Fchannels-integration.md` for event formats, configuration fields, and\nthe notification payload pattern used by `hooks\u002Fsession-end.mjs`.\n\n## Subagents (Conceptual Roles, Not Direct Dispatch Targets)\n\n**Important**: The Pokemon names below are **conceptual role labels**, not Agent tool `subagent_type` values. PDCA does NOT dispatch them directly via the Agent tool. Instead, the chained skills (`\u002Fscc:research`, `\u002Fscc:write`, `\u002Fscc:review`, `\u002Fscc:refine`) handle these roles internally — they map each role to the appropriate concrete subagent or model.\n\nIf you (the orchestrator) try to call `Agent(subagent_type: \"eevee\")`, it will fail. Always go through the chained skill, never bypass to a Pokemon role.\n\n```yaml\n# Plan dispatches two skills, so its roles split across them.\n# Roles handled inside \u002Fscc:research\nresearcher: { model: sonnet, role: eevee, purpose: \"collect sources with citations, minimum 5 distinct sources\" }\nanalyst: { model: sonnet, role: alakazam, purpose: \"list gaps, flag data conflicts, verify coverage\" }\n\n# Roles handled inside \u002Fscc:analyze — alakazam appears in both, pre-processing there rather than gap-listing\nanalyst: { model: sonnet, role: alakazam, purpose: \"structure raw research into patterns before a framework is applied\" }\nstrategist: { model: sonnet, role: mewtwo, purpose: \"apply the framework, cite evidence, challenge assumptions\" }\n\n# Role handled inside \u002Fscc:write\nwriter: { model: opus, role: smeargle, purpose: \"produce artifact from plan, skip-research skip-review\" }\n\n# Roles handled inside \u002Fscc:review (parallel dispatch)\ndeep-reviewer: { model: opus, role: xatu, purpose: \"thorough quality review with structured critic output\" }\ndevil-advocate: { model: sonnet, role: absol, purpose: \"attack weakest points, find logical gaps\" }\nfact-checker: { model: sonnet, role: porygon, purpose: \"verify claims against sources, flag unsupported statements\" }\ntone-guardian: { model: sonnet, role: jigglypuff, purpose: \"check voice and audience fit against .data\u002Fsoul\u002FSOUL.md Tone Rules and Anti-Patterns\" }\nstructure-analyst: { model: sonnet, role: unown, purpose: \"check organization, flow, format compliance\" }\n\n# Role handled inside \u002Fscc:refine\neditor: { model: opus, role: ditto, purpose: \"apply top 3 fixes per iteration, verify improvement\" }\n\n# Orchestration meta-role (the PDCA orchestrator itself)\npipeline-orchestrator: { model: sonnet, role: arceus, purpose: \"enforce gates, manage phase transitions, never skip phases, prefer domain hand-off over self-processing\" }\n```\n\n**How PDCA actually executes**: The orchestrator calls `\u002Fscc:research`, `\u002Fscc:write`, `\u002Fscc:review`, `\u002Fscc:refine` as Skill invocations. Each of those skills internally dispatches the right subagents (`general-purpose`, `code-reviewer`, etc.) using the Agent tool. PDCA never bypasses this layer.\n\n**Tier overrides are upward-only.** `--tier fable` raises the two spine roles — `writer` (smeargle) and `deep-reviewer` (xatu) — for high-stakes runs. Nothing lowers them: not a flag, and not an automatic fallback when a dispatch fails. If those dispatches are exhausted, escalate to the user instead of retrying at a cheaper tier, because every later phase inherits what the writer and the whole-document reviewer missed. To make a run cheaper, cut cycles or reviewers. See `references\u002Fdesign-principles.md` §10.\n\n## Gotchas\n\n- Do NOT skip gates. They prevent garbage-in-garbage-out.\n- Do NOT run Do without Plan unless user explicitly has source material ready.\n- Do NOT declare complete without Check phase verdict.\n- Do NOT route all Act findings to Loop — use the Action Router to classify root causes.\n- Full PDCA with deep research = significant token cost. Warn user at start.\n\nSee `gotchas.md` for extended gotchas (stuck detection, worktree lifecycle, permission transitions, session resume, automation mode).\n\n## MMBridge Session Handoff\n\nWhen mmbridge is detected (see `references\u002Fmmbridge-integration.md`) and the PDCA cycle exits with `APPROVED`, generate a handoff artifact.\n\n### When to Trigger\n\n- PDCA cycle reaches EXIT (APPROVED verdict from Check phase)\n- mmbridge is installed\n- Do NOT trigger on partial cycles or early exits\n\n### Command\n\n```bash\nmmbridge handoff --write .captures\u002Fpdca-handoff-${RUN_ID}.md\n```\n\n### Output\n\nThe handoff artifact summarizes:\n- What was researched (Plan phase sources and findings)\n- What was produced (Do phase artifact path and type)\n- What reviewers found (Check phase verdict and key findings)\n- What was refined (Act phase changes, if any)\n\nPresent the handoff path to the user:\n> \"PDCA cycle complete. Handoff artifact saved to `.captures\u002Fpdca-handoff-${RUN_ID}.md`\"\n\n### Usage\n\nThe handoff artifact serves two purposes:\n1. **New session briefing**: Load into a fresh Claude session for continuity\n2. **Team communication**: Share with colleagues as a work summary\n\nThis is an **optional convenience** — the PDCA cycle is complete regardless of whether the handoff succeeds.\n\n## MMBridge Embrace (Full-Cycle Acceleration)\n\nWhen mmbridge is detected, the PDCA orchestrator can optionally delegate the full cycle to `mmbridge embrace` for parallel multi-model acceleration.\n\n### When to Use\n\n- `--depth deep` with full PDCA: embrace provides multi-model research and debate in parallel with internal agents\n- Large-scope tasks where multi-model perspectives add value\n- NOT for `--depth shallow` or single-phase runs\n\n### Dispatch\n\nAt the start of a full PDCA cycle, if mmbridge is detected and depth is `deep`:\n\n```bash\nmmbridge embrace \"\u003Ctask summary>\" --json --non-interactive > \u002Ftmp\u002Fmmbridge-embrace-${RUN_ID}.json\n```\n\n- `--non-interactive`: auto-proceed through checkpoints (PDCA orchestrator manages gates, not embrace)\n- `--skip-phases`: use when specific PDCA phases should not have embrace acceleration (e.g., `--skip-phases security` if not a code project)\n\n### Merge Points\n\nEmbrace results are merged at multiple PDCA phases:\n- **Plan**: embrace research findings → supplemental sources for Eevee\n- **Check**: embrace review findings → additional voter in consensus gate\n- **Act**: embrace security findings → additional input for action router\n\n### Cost Note\n\n`mmbridge embrace` runs multiple models across multiple phases. Only activate at `deep` depth. Warn user about estimated cost at PDCA start.\n",{"name":5,"effort":38,"description":7,"user-invocable":34,"disable-model-invocation":39},"high",true,"main","3e2938e2698dc6fc652571205aad03aa1205ec3d","skills\u002Fpdca",9,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code","https:\u002F\u002Fgithub.com\u002Funclejobs-ai","2026-11-10 10:15:09",{"source":8,"skillName":5,"commands":48,"prompt":58,"zipAvailable":39},[49,52,55],{"tool":50,"command":51},"npx","npx skills add unclejobs-ai\u002Fsecond-claude-code --skill pdca",{"tool":53,"command":54},"bunx","bunx skills add unclejobs-ai\u002Fsecond-claude-code --skill pdca",{"tool":56,"command":57},"pnpm","pnpm dlx skills add unclejobs-ai\u002Fsecond-claude-code --skill pdca","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002FSKILL.md",[60,66,70,74,78,82,86,90,94,98,102,106,110,114],{"path":61,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":64,"sourceHtmlUrl":65},"gotchas.md","REFERENCE","参考文档",2858,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Fgotchas.md",{"path":67,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":68,"sourceHtmlUrl":69},"references\u002Fact-phase.md",12801,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fact-phase.md",{"path":71,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":72,"sourceHtmlUrl":73},"references\u002Faction-router.md",5042,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Faction-router.md",{"path":75,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":76,"sourceHtmlUrl":77},"references\u002Fchannels-integration.md",3617,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fchannels-integration.md",{"path":79,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":80,"sourceHtmlUrl":81},"references\u002Fcheck-phase.md",6311,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fcheck-phase.md",{"path":83,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":84,"sourceHtmlUrl":85},"references\u002Fcode-engineering-lane.md",3615,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fcode-engineering-lane.md",{"path":87,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":88,"sourceHtmlUrl":89},"references\u002Fdo-phase.md",16270,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fdo-phase.md",{"path":91,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":92,"sourceHtmlUrl":93},"references\u002Fdomain-pipeline-integration.md",4845,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fdomain-pipeline-integration.md",{"path":95,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":96,"sourceHtmlUrl":97},"references\u002Fobservability.md",6397,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fobservability.md",{"path":99,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":100,"sourceHtmlUrl":101},"references\u002Fphase-schemas.md",12283,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fphase-schemas.md",{"path":103,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":104,"sourceHtmlUrl":105},"references\u002Fplan-phase.md",10184,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fplan-phase.md",{"path":107,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":108,"sourceHtmlUrl":109},"references\u002Fquestion-protocol.md",1804,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fquestion-protocol.md",{"path":111,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":112,"sourceHtmlUrl":113},"references\u002Fstate-schema.md",4397,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fstate-schema.md",{"path":115,"resourceType":62,"resourceTypeDesc":63,"sizeBytes":116,"sourceHtmlUrl":117},"references\u002Fstuck-detection.md",8955,"https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fpdca\u002Freferences\u002Fstuck-detection.md",[119,130,144,156,167,179],{"skillNo":120,"name":121,"nameZh":6,"description":122,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":123,"categories":124,"keywords":125,"publishedAt":33},"01M6GNHTFEKCG6X5Q0VCH0T2C1","write","Use when producing newsletters, articles, reports, shorts, or social content","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fwrite\u002FSKILL.md",[],[121,126,127,27,128,129,30],"content","social content","reports","social",{"skillNo":131,"name":132,"nameZh":6,"description":133,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":134,"categories":135,"keywords":139,"publishedAt":33},"01M6GNHP58KN60BQ8T7QXCC43D","workflow","Use when chaining multiple \u002Fscc commands into a reusable workflow","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fworkflow\u002FSKILL.md",[136],{"categoryNo":137,"categoryName":138,"taxonomyType":21,"taxonomyTypeDesc":22},"01JBSEEDCAT000000000000023","Skill与智能体开发",[132,27,140,141,142,30,143,32],"chaining","reusable","commands","scc",{"skillNo":145,"name":146,"nameZh":6,"description":147,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":148,"categories":149,"keywords":150,"publishedAt":33},"01M6GNHHQZJG8AVTCYPYPFYTNE","translate","Use when translating content between English and Korean with formatting preservation","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Ftranslate\u002FSKILL.md",[],[146,27,151,152,30,153,154,155,32,126],"korean","formatting","preservation","english","with",{"skillNo":157,"name":158,"nameZh":6,"description":159,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":160,"categories":161,"keywords":162,"publishedAt":33},"01M6GNHD52AQE5YCZKRK6D71ZX","soul","Use when observing user patterns and synthesizing a persistent identity profile","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fsoul\u002FSKILL.md",[],[158,32,163,164,165,27,166,30],"identity","user","profile","patterns",{"skillNo":168,"name":24,"nameZh":6,"description":169,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":170,"categories":171,"keywords":172,"publishedAt":33},"01M6GNH8AX5EGA8BWF2JTK5HR0","Use when benchmarking content, strategy, or code against a quality bar with parallel specialized reviewers","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Freview\u002FSKILL.md",[],[24,173,174,175,126,176,177,178,27,30,155],"specialized","benchmarking","code","parallel","quality","strategy",{"skillNo":180,"name":25,"nameZh":6,"description":181,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":182,"categories":183,"keywords":185,"publishedAt":33},"01M6GNH4QFZP7H5KA12DR5PCNJ","Use when finding sources through iterative web exploration and synthesis","https:\u002F\u002Fgithub.com\u002Funclejobs-ai\u002Fsecond-claude-code\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002FSKILL.md",[184],{"categoryNo":19,"categoryName":20,"taxonomyType":21,"taxonomyTypeDesc":22},[25,186,27,187,188,30,189],"web","Synthesis","exploration","sources"]