claude-impl-tools

cmux 창 분할 기반 멀티-AI 합의 리뷰 엔진. 동일 워크스페이스에서 Gemini/Codex 패널을 나란히 분할해 3-Stage 파이프라인(의견 → 반론 → 합성)을 진짜 병렬로 실행합니다. multi-ai-review의 순차 실행 대신 두 AI가 동시에 리뷰하는 것을 실시간으로 볼 수 있음. "/cmux-ai-review", "cmux 코드 리뷰", "AI 패널 리뷰", "병렬 리뷰" 등 cmux 환경에서 여러 AI의 동시 리뷰가 필요한 요청이라면 반드시 이 스킬을 사용하세요.

cmux 창 분할 기반 멀티-AI 병렬 태스크 실행. 동일 워크스페이스에서 Codex(코드)/Gemini(디자인)/Claude(계획)를 패널로 분할해 동시에 실행합니다. multi-ai-run의 순차 실행과 달리 진짜 병렬. "/cmux-ai-run", "cmux로 AI 분업", "창 분할 병렬 실행", "codex gemini 동시 실행" 등 cmux 환경에서 여러 AI를 동시에 돌리고 싶은 요청이라면 반드시 이 스킬을 사용하세요.

cmux 기반 멀티-AI 팀 오케스트레이션. cmux 워크스페이스에 Claude/Gemini/Codex를 독립 프로세스로 배치해 물리적 다중 레벨 팀을 구성합니다. Agent Teams API 없이 진짜 병렬 실행. 이기종 AI 팀 구성, 사이드바 실시간 모니터링, 동적 팀 확장/축소 지원. "/cmux-orchestrate", "cmux로 팀 실행", "멀티 AI 팀", "gemini codex claude 병렬 실행" 등 cmux 환경에서 여러 AI를 팀으로 구성하는 요청이라면 반드시 이 스킬을 사용하세요.

cmux 터미널 멀티플렉서 제어 스킬. 워크스페이스/패널/서피스 관리, 명령 전송, 브라우저 자동화, 알림, 사이드바 메타데이터 제어. '/cmux', 'cmux', '워크스페이스 만들어', '패널 분할', '브라우저 열어', '알림 보내' 등 cmux 터미널 제어 요청이라면 반드시 이 스킬을 사용하세요.

Long context optimization with self-editing — extract key information, compress, summarize, and auto-prune stale context using the H2O + Context-1 patterns. Use this whenever documents are too long, the context window is running low, you need to synthesize multiple files, or accumulated tool results are polluting the context. Triggers immediately on 'compress this', 'summarize this', 'document is too long', 'context overflow', 'clean up context', 'prune context', '컨텍스트 정리'. Triggers on /compress, /prune.

Query and analyze test coverage. Use this to check coverage before modifying code, identify uncovered files before a PR, or track coverage trends over time. Triggers immediately on keywords like "coverage", "test pass", "uncovered areas", or "add tests".

Visualize the project dependency graph and detect circular dependencies. Use this before refactoring, during architecture reviews, before adding new domains, or when evaluating module separation. Trigger immediately when cross-domain coupling seems high or circular references are suspected. Triggers on /deps, "dependency graph", or "check circular dependencies".

Gray area 식별 및 사용자 결정 수집. CONTEXT.md를 생성하여 downstream 에이전트가 재질문 없이 작업 가능. GSD discuss-phase 패턴 기반.

Agile Sprint Master for regular user check-ins and sprint management. Use this for /agile, /sprint, starting a sprint, checkpoint requests, or any time you are running structured iterative development. Triggers on "start sprint", "checkpoint", "sprint status", "run task", or whenever you need to manage layered build progress across Skeleton → Muscles → Skin.

Query and visualize the full project architecture. Use this during project onboarding, before code reviews, before adding new domains, or whenever you need to understand the tech stack. Responds to questions like "show me the overall structure", "list all APIs", "what's the tech stack", or "explain the domain structure". Triggers on /architecture.

Query and analyze project change history and statistics. Use this before code reviews, when preparing for a deployment, when tracing change history for a specific file, or when analyzing activity by domain. Responds to questions like "recent changes", "change history for this month", or "list of bug fixes". Triggers on /changelog.

Immediate code review at task/PR completion. Auto-detects Git Diff, extracts TASKS.md context, runs Hook gates, and optionally invokes AI multi-review. Use this after completing a task, before committing, or whenever someone says "checkpoint", "review my code", "code review", or "check changes". Always trigger on task completion.

Sets up Agent Teams governance (team-lead + architecture/qa/design leads) for large projects, or a Mini-PRD for small ones. Use this skill at project kickoff, whenever a team structure is needed, or on requests like "set up governance", "configure project team", or "create agent team" — do not skip it.

Analyzes the blast radius and risk level before modifying any file. Run this skill before touching production code, before refactoring, and before changing core logic. It is mandatory for files under payment, auth, billing, and security paths. Triggers on /impact, "impact analysis", or "what does changing this file affect?"

Production system maintenance orchestrator. Automates change acceptance → impact analysis → safe modification → test reinforcement → change record in 5 ITIL-based stages. Use this skill for bug fixes, feature changes, config updates, incident response, and hotfixes — always, without exception.

Cross-project experience memory — your persistent knowledge base across sessions. Remembers what worked, what failed, and why, so every new session starts smarter. Log outcomes → recall lessons → apply patterns. Use when starting work that other projects have solved, when you need to avoid past mistakes, or when onboarding to a new project area. Also logs skill execution outcomes to improve future routing. Responds to: 'memento', 'cross-project knowledge', 'what did we learn', 'past experience', 'avoid repeating', 'session memory', 'global memory', '스킬 성능', '스킬 추천', '스킬 건강', '어떤 스킬이 좋아'.

Claude + Gemini CLI + Codex CLI multi-AI consensus engine. 3-Stage Pipeline (opinions → rebuttal → synthesis) drives consensus across domains: code review, market analysis, investment diligence, risk assessment, and more. Automatically detects domain from natural language and applies appropriate preset. Triggers: - "review this", "convene a council", "get multiple AI opinions", "deep review", "consensus" - "market analysis", "macro analysis", "investment review", "risk assessment" - Any situation needing multiple perspectives — trigger proactively

Routes each agent role to the optimal AI model (Claude/Gemini/Codex) and executes it. Automatically assigns Codex for code writing, Gemini for design, and Claude for planning and orchestration. Use this skill whenever you see "use Codex for code", "use Gemini for design", "use a different AI", or "model routing" requests. Triggers on /multi-ai-run.

Comprehensive quality audit before phase completion or deployment. Performs planning conformance, DDD validation, security checks, tests, browser verification, and quantitative metrics. Also enforces verification-before-completion discipline — no claims without evidence. Run this skill before deploying, before merging a PR, after completing significant changes, and whenever anyone claims 'it works' or 'tests pass'. Triggers immediately on 'quality check', 'pre-deploy check', 'audit this', 'QA', 'verify this', 'does this work', '품질 검사', '배포 전 검사'. Trigger: /audit, /evaluate, /verify.

Automatically detects and resumes incomplete work after CLI crashes, network drops, or agent errors. Use this skill whenever a session is interrupted, or on requests like "recover work", "interrupted task", "continue where you left off", or "what were you doing?"

OWASP Top 10 security vulnerability scanner. Detects secret exposure, injection flaws, auth/authz gaps, supply-chain risks, security misconfigurations, and logging failures via static analysis. Use this before every deployment, before every PR merge, and whenever auth/payment/billing code changes. Invoke immediately on "security check", "vulnerability scan", or "secret exposure" requests.

SI 프로젝트 요구사항 공학 스킬. 소크라테스식 요구사항 수집 → 도메인별 체크리스트 검증 → 산출물 자동 생성(요구사항분석서/기능정의서/화면정의서/추적표) → 변경 영향 분석. 대시보드 연동으로 고객 공유 지원.

Displays TASKS.md progress in real time on Line 3 of the Claude Code status bar. Use this whenever you want a live view of project progress or need task tracking at a glance. Guides "show progress", "set up statusline", or "task status" requests. Triggered by /statusline.

Interactively scaffolds a TASKS.md file from scratch. Use this at project start, whenever TASKS.md is missing, or whenever task structure is needed. Invoke immediately on "create tasks", "generate TASKS.md", "start project", or "organize to-dos" requests. Runs standalone. Triggered by /tasks-init.

Consolidates legacy task files (docs/planning/06-tasks.md, etc.) into the root TASKS.md. Use this whenever task files are scattered across multiple locations. Invoke immediately on "migrate tasks", "move 06-tasks to TASKS", or "consolidate tasks" requests.

Unified orchestration engine with 3 modes: team (Agent Teams API, 3-level hierarchy), auto (direct Task dispatch, worktree-based phases), and thin (ultra-minimal context for 50-200 tasks). Use this skill for any multi-task project execution — it auto-selects the right mode based on task count and infrastructure. Triggers on /team-orchestrate, /auto-orchestrate, /orchestrate, '에이전트 팀 실행', '오케스트레이트', '자동 실행', '완전 자동화', and any request to execute TASKS.md at scale.

File-based control plane that observes, explains, and controls the AI coding process. /whitebox is the sole product boundary — launch|status|explain|health|approvals expose run start, current state, intervention triggers, linked DEC evidence, approval control, and execution environment health. Use this to see what is running, why something is blocked, and to approve or reject gates.

Routes you to the right skill for any situation by auto-analyzing project state. Use this whenever you are unsure what to do next — it diagnoses the project and recommends the optimal skill. Invoke immediately on "what should I do?", "which skill?", "recommend a workflow", or "what's next?" questions. Triggered by /workflow.