원클릭으로
doc
Review production code and update all documentation under docs/ to match the current codebase state.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Review production code and update all documentation under docs/ to match the current codebase state.
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
Deep-dive analysis of an open-source repository. Clones the repo, explores architecture/patterns/internals, and produces up to 30 structured Markdown files in analysis/<repo-name>/. Use when analyzing a GitHub repo for learnings, comparison, or adoption roadmap.
Review code changes for quality, security, and correctness. Use this before merging PRs, after significant refactors, or when reviewing agent output.
Review and optimize LLM API costs in the orchestrator. Analyze model routing, budget tracking, prompt caching, and retry logic for cost efficiency.
Deploy services using Docker/OrbStack. Use this to build and deploy containers, run health checks, and manage deployments.
Build and manage Docker containers via OrbStack. Use this to build images, run containers, check service health, and manage docker-compose services.
End-to-end feature development — implement, iterate with user feedback, test, SOLID review, document, commit & push. Use when building a new feature or making a significant code change that needs quality assurance.
| name | doc |
| description | Review production code and update all documentation under docs/ to match the current codebase state. |
| allowed-tools | Bash, Read, Edit, Write, Grep, Glob, Agent |
| user-invocable | true |
Review the entire production codebase and bring all docs under docs/, docs/website/docs/, CLAUDE.md, and README.md in sync with the actual code.
Optional argument: a hint string suggesting which area to focus on (e.g., "dashboard API", "graph engine", "deployment"). If provided, prioritize that area but still do a full pass.
STOP on any failure — do not continue if a phase fails.
Read the actual codebase to understand what exists right now:
CLAUDE.md (the canonical project reference)src/agent_orchestrator/core/ — list all modules, public classes, and key functionssrc/agent_orchestrator/dashboard/ — list endpoints, WebSocket handlers, event typessrc/agent_orchestrator/providers/ — list all providerssrc/agent_orchestrator/skills/ — list all skillsdocker/, docker-compose*.yml — list all services and their configterraform/ — list all modules.claude/agents/ — list all agents by category.claude/skills/ — list all skillstests/ — count test files and total test classesBuild a mental map of: modules, abstractions, APIs, services, agents, skills, and their relationships.
If the user provided a focus hint, note which docs are most relevant to that area.
For each documentation file, compare its content against the source-of-truth map. Check for:
Always audit (these are the primary targets):
| File | What to check |
|---|---|
CLAUDE.md | Project structure, modules, abstractions, agents, skills, dashboard sections |
README.md | Overview, quick start, features list |
docs/architecture.md | Core abstractions, patterns, module relationships |
docs/components.md | Dashboard components, API endpoints, sequence diagrams |
docs/deployment.md | Docker services, deploy steps, environment variables |
docs/security.md | Auth, RBAC, network, secrets |
docs/cache-strategy.md | Cache integration points, implementation status |
docs/roadmap.md | Completed vs planned features |
Also audit (website mirror — should match docs/):
| File | Mirror of |
|---|---|
docs/website/docs/architecture/overview.md | docs/architecture.md |
docs/website/docs/architecture/components.md | docs/components.md |
docs/website/docs/architecture/agents.md | Agent structure |
docs/website/docs/architecture/skills.md | Skills list |
docs/website/docs/architecture/providers.md | Provider list |
docs/website/docs/architecture/graph-engine.md | Graph/StateGraph docs |
If the user provided a focus hint, audit those docs first and in more detail.
Before making any changes, produce a DOC AUDIT REPORT:
DOC AUDIT REPORT
================
Focus: <user hint or "full review">
Files audited: <count>
STALE (needs update):
- <file>: <what's wrong>
- ...
GAPS (missing documentation):
- <topic>: should be documented in <file>
- ...
INACCURATE (contradicts code):
- <file>: <what's wrong>
- ...
UP TO DATE:
- <file>: OK
- ...
Use the Agent tool with subagent_type=Explore for deep codebase exploration when needed.
For each finding from Phase 3:
docs/website/docs/), update both.claude/agents/ and .claude/skills/src/agent_orchestrator/core/docker-compose*.ymlAfter all edits:
# Verify agent count in CLAUDE.md matches actual agents
find .claude/agents -name "*.md" -not -path "*/team-lead.md" | wc -l
grep -c "agents" CLAUDE.md || true
# Spot-check: extract file paths from docs and verify they exist
grep -ohP 'src/agent_orchestrator/\S+\.py' docs/*.md CLAUDE.md | sort -u | while read f; do
[ ! -f "$f" ] && echo "MISSING: $f"
done
git status and git diff --statgit commit -m "$(cat <<'EOF'
docs: sync documentation with current codebase
<summary of what was updated>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
EOF
)"
git push
DOC SYNC REPORT
===============
Focus: <user hint or "full review">
Files Audited: <count>
Files Updated: <count>
Files OK: <count>
Changes:
- <file>: <what was fixed>
- ...
Commit: <hash> <message>
Push: [OK]