mise-en-place（就位）

Mise-en-Place

Turn a raw idea or existing spec into a detailed, implementation-ready prd.json for ratatouille.

Trigger Formats

• mise-en-place: <short idea>
• mise-en-place: <file path> (e.g., mise-en-place: SPEC.md)

Inputs

• Idea prompt OR spec file
• Optional: existing codebase to inspect for stack + constraints

Outputs

• prd.json (root)

Workflow

1) Intake

1. If user provides a file path, read it. If missing, ask for the correct path.
2. If user provides a prompt, treat it as the seed spec.
3. If a codebase exists, do a quick scan for stack + constraints (configs, lockfiles) to avoid redundant questions.

2) Interview + Brainstorm

Ask focused questions to fill gaps. Keep it fast and structured. If the user prefers brevity, batch 3-5 questions at a time.

Minimum areas to cover:

• Problem + target users
• Core user flow (start → success)
• Features (must-have vs nice-to-have)
• Non-goals/out of scope
• Tech stack expectations (or confirm detected stack)
• Data model + APIs/integrations
• Security/privacy requirements
• Design direction (if UI)
• Constraints (deadline, budget, compliance)

3) Research (Web)

Run targeted web research for:

• Domain-specific standards or best practices
• Chosen stack + library gotchas
• Security/compliance requirements (if applicable)
• Competitive benchmarks or common UX patterns

Summarize key findings (short bullets) and capture sources in prd.json.

4) Generate prd.json

Create a single, canonical JSON file in the repo root:

{
  "meta": {
    "title": "",
    "created": "YYYY-MM-DD",
    "updated": "YYYY-MM-DD",
    "source": "mise-en-place",
    "seed": "prompt|file"
  },
  "vision": {
    "problem": "",
    "audience": "",
    "success": ""
  },
  "scope": {
    "in": [""],
    "out": [""]
  },
  "user_flow": [
    "Discovery",
    "Onboarding",
    "Core loop",
    "Exit/return"
  ],
  "features": [
    {
      "id": "F-001",
      "name": "",
      "description": "",
      "inputs": [""],
      "outputs": [""],
      "edge_cases": [""]
    }
  ],
  "requirements": {
    "functional": [""],
    "non_functional": [""]
  },
  "data_model": [
    {
      "entity": "",
      "fields": [""],
      "relationships": [""]
    }
  ],
  "integrations": [
    {
      "name": "",
      "purpose": "",
      "constraints": ""
    }
  ],
  "design": {
    "style": "",
    "references": [""]
  },
  "tech_stack": {
    "frontend": "",
    "backend": "",
    "db": "",
    "auth": "",
    "hosting": "",
    "realtime": "",
    "other": [""]
  },
  "risks": [""],
  "research": [
    {
      "topic": "",
      "summary": "",
      "sources": [""]
    }
  ],
  "tasks": [
    {
      "id": "T-001",
      "title": "",
      "goal": "",
      "scope": "one iteration",
      "acceptance_criteria": [""],
      "tracer_bullet": "thin end-to-end slice",
      "dependencies": [""],
      "status": "todo",
      "notes": ""
    }
  ]
}

5) Task Decomposition Rules

• Each task must be one ratatouille iteration.
• Provide a tracer-bullet path for each task.
• Keep tasks independent where possible.
• Include explicit acceptance criteria.

6) Handoff Hint

Finish by telling the user to run ratatouille:

• ratatouille: prd.json

Rules

• Be concise and precise.
• Do not implement code.
• Store everything in prd.json.