메뉴
Skill navigator. List active skills, locate archived ones, choose the right sequence per scenario.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Enforces format rules + Pre-Creation Protocol for Skill, Sub-agent, Rule files. TRIGGER: create any of them. NOT FOR: WHETHER to create (use skill-author), registration (use skill-graph-manager).
Governs WAL fragment writes (domain/api/rules/data/architecture) at Archive. TRIGGER: STANDARD Archive, dimensions elected via h-archive. NOT FOR: PATCH; wiki authoring (use documentation-curator).
Runs one-round adversarial critique. TRIGGER: HIGH-risk STANDARD — Explorer (Cat A), Propose/Review (Cat B/C). NOT FOR: code review (use code-reviewer); PATCH; >1 round per category.
Orchestrate the repo-aligned AI engineering pipeline (plan → decisions → eval → improve → cleanup → verify → archive) with explicit approval gates. TRIGGER when user asks to run the full pipeline or go from idea to delivery; DO NOT TRIGGER for small tasks or when user invokes a specific phase.
Decides whether to author a SKILL.md. TRIGGER: 'create a skill', recurring workflow without one, broken trigger. NOT FOR: one-offs, rules, non-skill artifacts (use authoring-standards).
Verifies each AC against implementation in a single pass before Archive. TRIGGER: Phase 5 QA with AC ≤ 3 AND risk ≠ HIGH. NOT FOR: AC ≥ 4 or HIGH risk (use ultraqa); writing tests (use lead-engineer).
| name | skill-index |
| description | Skill navigator. List active skills, locate archived ones, choose the right sequence per scenario. |
This framework keeps 29 "active" skills auto-loaded by Claude Code. 13 lower-frequency skills are stored under .claude/skills-archive/ and need to be re-activated on demand (see § Archive).
Scope: A typical Claude Code install may surface external skills (lark-*, claude-api, loop, schedule, etc.). Those are global utilities — ignore unless asked by name.
| Scenario | Skill sequence |
|---|---|
| Any non-trivial input (PRD / bug / signal) | input-classifier → route below |
| Idea / Feedback / Compliance input (STANDARD) | input-classifier → ambiguity-gatekeeper → (if PASS) requirement-engineer |
| Writing a new feature | brainstorming → task-decomposition-guide → java-architecture-standards → test-driven-development → ultraqa → wal-documentation-rules → remember |
| Processing a PRD | product-manager-expert (Ingestion) → task-decomposition-guide → feature flow |
| Fixing a bug | root-cause-debug → test-driven-development → ac-verify |
| Code review / QA (PATCH/LOW) | code-review-checklist → java-testing-standards → ac-verify |
| Code review / QA (STANDARD/MEDIUM+) | code-reviewer sub-agent → java-testing-standards → ultraqa |
| Security or HIGH risk change | security-review-checklist → adversarial-review → code-reviewer sub-agent → ac-verify |
| Cleanup after AI-heavy session | ai-slop-cleaner → code-review-checklist |
| Recording a design decision | architecture-decision-records |
| Knowledge preservation | wal-documentation-rules (Archive) → remember (cross-session) |
| Pre-Explorer codebase context | local-code-intelligence (BM25 + symbol index + failure memory) |
| Migration / Greenfield / Incident / EPIC / PRD / Release / Pipeline | See .claude/rules/lifecycle.md Special Scenarios — the matching scenario inlines the archive path to read |
These 29 skills live under .claude/skills/<name>/SKILL.md and are visible to the Skill tool without further action.
| Skill | Lifecycle Phase(s) | Primary Agent |
|---|---|---|
| brainstorming | Explorer / Propose | Requirement Engineer |
| task-decomposition-guide | Propose / Review | System Architect |
| impl-plan | Propose / Implement | System Architect / Lead Engineer |
| root-cause-debug | Implement / QA | Lead Engineer / Code Reviewer |
| test-driven-development | Implement | Lead Engineer |
| ac-verify | QA / Archive | Code Reviewer / Knowledge Extractor |
| code-review-checklist | QA | Code Reviewer |
| wal-documentation-rules | Archive | Knowledge Extractor |
| skill-graph-manager | Any (skills change) | Main agent (inline) |
| java-architecture-standards | Propose / Implement | System Architect / Lead Engineer |
| java-coding-style | Implement | Lead Engineer |
| java-testing-standards | QA | Code Reviewer |
| mybatis-sql-standard | Propose / Implement | System Architect / Lead Engineer |
| Skill | Required When |
|---|---|
| cognitive-bias-checklist | Requirement Engineer / System Architect — Propose phase |
| spec-quality-checklist | Documentation Curator / pre-gate self-check |
| decision-frameworks | System Architect — ambiguous root cause or design choice |
| ultraqa | QA phase, AC count ≥ 4 OR HIGH risk |
| security-review-checklist | HIGH risk change touching auth/data/secrets |
| skill-author | Decide whether a new SKILL.md is warranted; drives the authoring pre-flight |
| authoring-standards | Once "yes", provides format rules for Skill / Sub-agent / Rule artifacts + Pre-Creation data-sufficiency gate |
| skill-index | This file |
These were moved back from archive because they fit the daily flow.
| Skill | Use When |
|---|---|
| adversarial-review | HIGH risk Review phase — one isolated round of critique. Required by lifecycle.md HIGH flow. Detects requirements/design contradictions. |
| stakeholder-conflict-resolver | Downstream from adversarial-review Category A CRITICAL — when conflicting requirements come from different stakeholders (PM / frontend / security / legal), produces structured conflict map + resolution decision. |
| local-code-intelligence | Explorer phase — BM25 wiki search + Java symbol index + failure memory. Run before reading source files. |
| remember | Archive phase — classify discovered knowledge into project memory / notepad / docs. |
| ai-slop-cleaner | After AI-heavy session — regression-safe cleanup of dead code, duplicates, over-abstraction. |
| architecture-decision-records | When an architectural decision is made — capture as ADR. Pairs with HIGH-risk ≥2-ADR requirement. |
| input-classifier | Front door for any non-trivial input (PRD, bug report, signal, security finding) before routing. |
| product-manager-expert | PRD generation (Mode A) or PRD ingestion → AC + implementation queue (Mode B). |
| Phase | Agent | Skills |
|---|---|---|
| Pre-Explorer | — | input-classifier (if input is non-trivial) |
| Explorer | Requirement Engineer | local-code-intelligence → input-classifier → (ambiguity-gatekeeper if Idea/Feedback/Compliance/Security) → brainstorming → (cognitive-bias-checklist) → (spec-quality-checklist) |
| Propose / Review | System Architect | brainstorming (one ADR per actual irreversible decision via architecture-decision-records; zero ADRs allowed with explicit "mechanical" note) → task-decomposition-guide → decision-frameworks |
| Review (HIGH only) | Devil's Advocate | adversarial-review (one isolated round) |
| Implement | lead-engineer (scope_guard.py hook enforces Allowed Scope) | impl-plan → java-architecture-standards / java-coding-style / mybatis-sql-standard → root-cause-debug / test-driven-development |
| QA | Code Reviewer | code-review-checklist → java-testing-standards → ultraqa → (security-review-checklist) |
| Cleanup | Lead Engineer | (optional) ai-slop-cleaner |
| Archive | Knowledge Extractor | wal-documentation-rules → ac-verify → remember (cross-session lessons) |
For Greenfield / Migration / EPIC / Incident / Pipeline / Release flows, the matching scenario in .claude/rules/lifecycle.md inlines the archive path the agent must read.
The following 12 skills live under .claude/skills-archive/<name>/SKILL.md. They are not auto-injected — instead, each is referenced inline by the rule or agent that needs it, with the full path written in place. No central lookup table, no "decide whether to activate" step.
| Skill | Referenced from |
|---|---|
incident-response | .claude/rules/lifecycle.md → Scenario A |
migration-planner | .claude/rules/lifecycle.md → Scenario B2 (mutating DDL / migration); B1 additive is PATCH and skips this skill |
greenfield-scaffold | .claude/rules/lifecycle.md → Scenario GREENFIELD |
blueprint | .claude/rules/lifecycle.md → Scenario EPIC |
dispatching-parallel-agents | .claude/rules/lifecycle.md → Scenario EPIC |
using-git-worktrees | .claude/agents/lead-engineer.md → HIGH-risk / parallel work |
ai-pipeline | .claude/rules/lifecycle.md → Scenario PIPELINE |
self-improve | .claude/rules/lifecycle.md → Scenario PIPELINE |
eval-harness | .claude/rules/lifecycle.md → Scenario PIPELINE |
external-research | .claude/rules/lifecycle.md → Scenario D + Scenario PIPELINE |
release | .claude/rules/lifecycle.md → Scenario RELEASE |
deepinit | .claude/rules/lifecycle.md → Scenario GREENFIELD |
linter-severity-standard | Any gate script invocation — OK/WARN/FAIL exit-code contract reference |
Each referenced location writes the full .claude/skills-archive/<name>/SKILL.md path inline. When the rule fires, the agent reads that exact file — no judgment about whether to mount it, no lookup needed.
Auto-loading 12 rarely-used skill descriptions costs ~1,800 tok on every session. By keeping these out of .claude/skills/ and inlining their paths at the point of use, the framework retains the full capability surface while paying the cost only when the rule actually fires.
.claude/agents/ — agent catalog; each agent file lists which skills it depends on.claude/rules/lifecycle.md — when each lifecycle phase fires.claude/skills/skill-author/SKILL.md — used when adding a new skill (will prompt you to register here).claude/skills/authoring-standards/SKILL.md — format authority for Skill / Sub-agent / Rule files; enforces Pre-Creation Protocol