---

name: breakdown-analyze-prd description: Analyze a PRD to extract features, tech stack, and implementation requirements. Called by /breakdown skill during Phase 2. context: fork allowed-tools: Read Glob Grep model: claude-haiku-4-5

PRD Analysis

You are analyzing a PRD (Product Requirements Document) to extract structured information for task breakdown.

Input

The calling skill will provide the full PRD XML content.

Your Task

Extract and structure the following information from the PRD:

1. Features

For each feature in <features>:

• Name
• Priority (must-have, should-have, could-have, won't-have)
• Summary
• File path to detailed spec (if exists)

2. Tech Stack

From <tech-stack>:

• Project type: Extract from <type> element (greenfield/brownfield)
• Project path: Extract from <project-path> element (brownfield only)
• All technologies with versions from <selected> element
• Template path: Detect from technologies (e.g., "Python + FastAPI" → webapps/backends/python)
• Rationale for choices from <rationale> element

3. Data Models (Inferred)

Analyze features to identify implied database models:

• Model name
• Key fields (inferred from feature descriptions)
• Relationships between models

4. API Endpoints (Inferred)

Analyze features to identify implied API endpoints:

• HTTP method
• Path
• Purpose
• Related feature

5. Frontend Components (Inferred)

Analyze features and UI layout to identify:

• Component name
• Type (page, layout, widget, form, etc.)
• Related feature

6. Dependencies

From <dependencies>:

• Name
• Version
• Purpose

7. Template Information

If tech-stack references a template:

• Template path
• What the template provides (auth, database setup, etc.)

Output Format

Return a JSON object with this structure:

Copy{
  "prd_meta": {
    "name": "Project Name",
    "slug": "project-slug",
    "status": "complete|in-progress"
  },
  "features": [
    {
      "name": "Feature Name",
      "priority": "must-have",
      "summary": "Brief description",
      "spec_file": "features/feature-slug.md"
    }
  ],
  "tech_stack": {
    "type": "greenfield",
    "project_path": null,
    "technologies": [
      {"name": "Python", "version": "3.11", "purpose": "Backend language"},
      {"name": "FastAPI", "version": "0.100+", "purpose": "API framework"}
    ],
    "template": {
      "path": "webapps/backends/python",
      "provides": ["auth", "database", "docker"],
      "detected_from": "Python + FastAPI in selected technologies"
    }
  },
  "data_models": [
    {
      "name": "Project",
      "fields": ["id", "name", "slug", "status", "content"],
      "relationships": ["has_many: Conversation"]
    }
  ],
  "api_endpoints": [
    {
      "method": "POST",
      "path": "/api/projects",
      "purpose": "Create new project",
      "feature": "Project Management"
    }
  ],
  "frontend_components": [
    {
      "name": "Dashboard",
      "type": "page",
      "feature": "Dashboard"
    }
  ],
  "dependencies": [
    {
      "name": "anthropic",
      "version": "latest",
      "purpose": "LLM API client"
    }
  ]
}

Template Detection

Detect the template from <tech-stack><selected> text using these mappings:

If template is explicitly mentioned in the PRD (e.g., "from webapp template"), use that. If no match found, set template.path to null and note in detected_from.

Analysis Guidelines

1. Be thorough: Capture all features, not just the obvious ones
2. Infer carefully: Data models and APIs should be reasonable inferences, not guesses
3. Use PRD language: Match names and terminology from the PRD
4. Note uncertainties: If something is unclear, include it with a note
5. Brownfield: If <type>brownfield</type>, expect <project-path> to be present

Example Inference

Given feature:

Copy<feature priority="must-have">
  <name>Project Management</name>
  <summary>Create, switch, and list PRD projects with status visibility</summary>
</feature>

Infer:

• Data model: Project with fields (id, name, status, created_at)
• API endpoints: GET /projects, POST /projects, GET /projects/:id, PUT /projects/:id
• Frontend: ProjectList component, ProjectCard component

Do NOT

• Make up features not in the PRD
• Guess at specific implementation details
• Include your own opinions about architecture
• Truncate or summarize features

Return the complete JSON analysis.