plan

name	plan
description	Use when the user needs a structured implementation plan with concrete TODOs. Triggers on "플랜", "계획", "plan", or after brainstorming completes.

Plan — Structured Implementation Planning

Overview

Create an implementation plan with concrete, agent-executable TODOs. The plan must be detailed enough that an executor agent can complete each TODO without human intervention.

Process

1. Gather Context

Delegate to explore agent (standard category):

• Read existing codebase structure
• Identify files that will be affected
• Find existing patterns and conventions

If a design document exists (from brainstorming), read it first.

2. Interview (if needed)

If requirements are unclear, ask the user — one question at a time, prefer multiple choice.

Only ask about user preferences, not technical details you can discover from the codebase:

• "테스트 프레임워크는 기존 프로젝트에서 사용하는 걸 그대로 쓸까?"
• "API 엔드포인트 이름은 어떤 컨벤션으로?"

Skip the interview if requirements are already clear from the design document.

3. Create Plan

Delegate to planner agent (standard or complex category).

TODO Structure

Each TODO must contain:

### TODO N: [Short title]
- **What**: Exactly what to do (1-2 sentences)
- **File(s)**: Exact file paths to create or modify
- **Tags**: [frontend | ui | backend | config | test] (1개 이상 필수)
- **Must NOT**: Explicit constraints
- **Acceptance**: How to verify completion
- **QA**: Concrete test scenario (exact inputs → expected outputs)

Plan Rules

1. Each TODO = 2-5 minutes of executor agent work
2. Every TODO references specific file paths
3. Acceptance criteria must be agent-executable (no human judgment needed)
4. QA scenarios use concrete data ("[e-mail]", not "[email]")
5. Dependencies between TODOs are explicitly noted
6. Independent TODOs are marked for potential parallel execution
7. Actively identify indirect dependencies between TODOs — if their target files are connected by import/export relationships, mark them as dependent

4. Self-Review

Before presenting to the user, verify:

• All TODOs have concrete acceptance criteria
• File references exist or are clearly marked as "to be created"
• No unvalidated assumptions
• Scope is clear (in AND out)
• No YAGNI violations

5. Save

After user approval, the planner agent saves the plan directly to vault using vault-save.sh via Bash tool:

1. Write plan content to a temp file
2. Run: bash scripts/vault-save.sh --project "{PROJECT}" --type "plan" --title "{TITLE}" --file "$tmp_file"
3. Handle response:
   • CREATED:{path} → extract vault path. Return this path to the orchestrator.
   • DUPLICATE:{path} → ask user whether to update or skip
   • Error (exit code != 0) → return error message to orchestrator
4. Clean up temp file

WARNING: NEVER include [Plan] prefix in --title — vault-save.sh adds it automatically.

Optional: Plan Review

If the plan is complex or the user requests it, invoke agmo:plan-review for a planner ↔ critic validation loop.

Transition to Execution

After the plan is approved and saved to vault:

• User says "실행해줘" / "구현해줘" / "시작해줘" → invoke agmo:execute with the vault path: Skill(skill="agmo:execute", args="--plan-path {VAULT_PATH}")
• The vault path format is an absolute path, e.g.: agmo-everywhere/plans/[Plan]-제목.md
• Execute automatically links ralph for persistent completion