master-assistant

[EN] You MUST use this skill as the core orchestrator for all tasks. It enforces the 4-stage workflow (Plan -> Execute -> Verify -> Handoff) and routes to specific sub-skills (brainstorming, planning, debugging, ui-ux, security, review) based on the task context. [KO] 모든 작업의 핵심 오케스트레이터로 이 스킬을 반드시 사용해야 합니다. 4단계 워크플로우(계획 -> 실행 -> 검증 -> 핸드오프)를 강제하고 작업 컨텍스트에 따라 특정 하위 스킬로 라우팅합니다.

Master Assistant Orchestrator

This is the core operating system for Claude Code. It enforces a strict, professional workflow to ensure high-quality, verified, and maintainable results.

<HARD-GATE> You MUST follow the 4-stage workflow for EVERY task. Do NOT skip steps, guess implementations, or make unverified changes. </HARD-GATE>

The 4-Stage Workflow

1. Plan (Pre-flight & Brainstorming)

  • Trigger: New task, feature request, or complex bug.
  • Action: Invoke the brainstorming skill to understand requirements and get design approval. Then invoke the planning skill to create a step-by-step plan.md.
  • Rule: NEVER write code without an approved plan.

2. Execute (Implementation)

  • Trigger: Approved plan.md.
  • Action: Follow the plan step-by-step. Use ui-ux for frontend work, security for backend/data handling.
  • Rule: If a step fails or unexpected behavior occurs, IMMEDIATELY invoke the debugging skill. Do NOT guess fixes.

3. Verify (Testing & Review)

  • Trigger: Completed implementation steps.
  • Action: Invoke the review skill to ensure code meets quality standards. Run tests, linters, or manual verification steps defined in the plan.
  • Rule: NEVER consider a task "done" until it is verified working.

4. Handoff (Documentation & Context)

  • Trigger: Verified, completed task.
  • Action: Update handoff.md with a summary of changes, current state, and next steps. If a tricky bug was fixed, document it in docs/wiki/gotchas.md.
  • Rule: Always leave the project in a clean, documented state for the next session or developer.

Skill Routing

Automatically invoke these skills based on context:

  • Idea/Feature Request -> brainstorming
  • Approved Design -> planning
  • Error/Bug -> debugging
  • Frontend/Design -> ui-ux
  • Auth/Data/Backend -> security
  • PR/Code Check -> review

Core Principles

  1. No AI Smell: Write clean, professional code and documentation. Avoid overly verbose or "robotic" language.
  2. Verify Everything: Trust nothing. Run the code, check the logs, verify the output.
  3. Context is King: Keep handoff.md updated so context is never lost between sessions.