prd-breakdown

PRD Breakdown - Analyze and Decompose Product Requirements

Analyze a Product Requirements Document (PRD) and decompose it into atomic features that integrate with the claude-harness workflow.

Arguments: $ARGUMENTS

Phase 0: PRD Input Detection & Storage

1. Detect PRD source (in priority order):

   • If arguments start with @ -> treat as file reference (e.g., @./docs/prd.md)
   • Else if --url flag provided -> fetch from GitHub issue
   • Else if --file flag provided -> read from specified file path
   • Else if file ./.claude-harness/prd.md exists -> read from file
   • Else if arguments provided -> treat as inline PRD markdown
   • Else -> prompt user for interactive input

2. Validate PRD format:

   • Check minimum length (at least 100 characters of content)
   • If Markdown: verify structure (sections, requirements)
   • If plain text: parse as-is
   • If too large (>100KB): warn user, ask to focus on specific sections

3. Store PRD input:

   • Create .claude-harness/prd/ directory if missing
   • Save PRD content to .claude-harness/prd/input.md
   • Create .claude-harness/prd/metadata.json:

     {
       "version": 1,
       "sourceType": "inline|file|github|interactive",
       "fetchedAt": "{ISO timestamp}",
       "sourceUrl": "{URL or path}",
       "hash": "{SHA256 of PRD}",
       "characterCount": 0,
       "sections": 0
     }
     

---

Phase 1: PRD Analysis

Perform comprehensive analysis from three perspectives:

4. Product Analysis:

   • Extract business goals, user personas, functional requirements
   • Identify non-functional requirements, dependencies, constraints

5. Architecture Analysis:

   • Review feasibility and technical complexity
   • Propose implementation order (dependency graph)
   • Identify risks and mitigations
   • Suggest MVP features

6. QA Analysis:

   • Define acceptance criteria for each requirement
   • Identify edge cases and error scenarios
   • Specify performance/security requirements

7. Save analysis results to .claude-harness/prd/analysis.json:

   {
     "version": 1,
     "analyzedAt": "{timestamp}",
     "product": {
       "businessGoals": [...],
       "userPersonas": [...],
       "functionalRequirements": [...]
     },
     "architecture": {
       "feasibilityAssessment": [...],
       "implementationOrder": [...],
       "mvpFeatures": [...],
       "dependencies": {...}
     },
     "qa": {
       "verificationFramework": {...},
       "edgeCases": [...]
     }
   }
   

Phase 2: Breakdown Generation

8. Transform analysis into atomic features:

   • For each functional requirement (from product analysis):
     • Generate feature name (readable title)
     • Extract acceptance criteria (from QA analysis)
     • Determine complexity from architecture assessment
     • Identify dependencies
     • Assign risk level

9. Resolve dependencies:

   • Build dependency graph: feature A depends on B, B depends on C
   • Topologically sort (ensures dependencies implemented first)
   • Detect cycles: ERROR if circular dependency found
   • Generate priority ordering

10. Generate feature specifications (with structured Gherkin acceptance criteria):

    {
      "id": "feature-XXX",
      "prdSource": {
        "section": "Section Name",
        "requirement": "R001"
      },
      "name": "Feature Title",
      "description": "One-line description",
      "detailedDescription": "Full description from PRD",
      "priority": 1,
      "dependencies": ["feature-YYY"],
      "acceptanceCriteria": [
        {
          "scenario": "Descriptive scenario name",
          "given": "precondition (context setup)",
          "when": "action performed",
          "then": "expected outcome"
        }
      ],
      "riskLevel": "low|medium|high",
      "estimatedComplexity": "low|medium|high",
      "mvpFeature": true|false
    }
    

    Important: acceptanceCriteria MUST use structured Gherkin format ({ scenario, given, when, then }) -- not plain strings. This enables the ATDD workflow in /flow --team where the tester teammate programmatically iterates scenarios to write executable tests.

11. Apply limits (if --max-features N provided):

    • Sort by priority, keep top N
    • Summarize excluded features

Phase 3: Feature Review & Creation

12. Generate preview showing (skip if --auto flag provided):

    • Total PRD sections analyzed
    • Functional requirements extracted
    • Features to create (grouped by priority)
    • MVP features highlighted
    • Risk assessment summary

    PRD BREAKDOWN ANALYSIS COMPLETE
    Sections: 5 | Requirements: 23 | Features: 8
    MVP Features: 3 | High-Risk: 1 | Dependencies: 5
    
    FEATURES (by priority):
      1. [MVP] Add user authentication
         Risk: MEDIUM | Complexity: MEDIUM | No dependencies
    
      2. Build user dashboard
         Risk: LOW | Complexity: LOW | Depends on: #1
      ... (6 more)
    
    Create features? [Y/n/select/review]
    

13. Handle user response:

    • Y: Create all features (go to step 14)
    • n: Stop here, show file path: .claude-harness/prd/breakdown.json
    • select: Show multi-select menu, create only selected features
    • review: Display full breakdown details for inspection

14. Create features in .claude-harness/features/active.json:

    • For each selected feature:
      • Generate next sequential feature ID (read active.json, find max, increment)
      • Add feature entry with full PRD metadata:

        {
          "id": "feature-XXX",
          "name": "...",
          "description": "...",
          "priority": N,
          "status": "pending",
          "acceptanceCriteria": [
            {
              "scenario": "...",
              "given": "...",
              "when": "...",
              "then": "..."
            }
          ],
          "prdMetadata": {
            "section": "...",
            "breakdown": "prd-{date}-{hash}"
          },
          "verification": {
            "build": "{auto-detected}",
            "tests": "{auto-detected}",
            "lint": "{auto-detected}",
            "typecheck": "{auto-detected}"
          },
          "relatedFiles": [],
          "github": {
            "issueNumber": null,
            "prNumber": null,
            "branch": "feature/feature-XXX"
          },
          "createdAt": "{timestamp}",
          "updatedAt": "{timestamp}"
        }
        

Phase 3.5: GitHub Issue Creation (unless --no-issues)

GitHub issues are created by default for all generated features. Skip with --no-issues.

15. Pass 1 -- Create issues with rich bodies (in dependency order -- dependencies first):

    • Parse GitHub owner/repo from git remote get-url origin (or use cached from SessionStart)

    • For each created feature (ordered so dependencies are created first):

      1. Build labels:

         • Base: ["feature", "prd-generated", "claude-harness"]
         • If mvpFeature is true: add "mvp"
         • If riskLevel is "high": add "high-risk"

      2. Build rich issue body:

         ## Description
         
         {feature.detailedDescription or feature.description}
         
         ### PRD Source
         **Section:** {prdSource.section} | **Requirement:** {prdSource.requirement}
         
         ---
         
         ## Acceptance Criteria
         
         **Scenario: {scenario}**
         - **Given** {given}
         - **When** {when}
         - **Then** {then}
         
         {repeat for each criterion in acceptanceCriteria}
         
         ---
         
         ## Implementation Context
         
         | Attribute | Value |
         |-----------|-------|
         | Priority | {priority} |
         | Complexity | {estimatedComplexity} |
         | Risk Level | {riskLevel} |
         | MVP Feature | {mvpFeature ? "Yes" : "No"} |
         
         ### Related Files
         {for each file in relatedFiles: "- `{file}`"}
         {if empty: "- To be determined during planning"}
         
         ### Implementation Hints
         {from architecture analysis for this feature: suggested approach, tech stack considerations, risks and mitigations}
         
         ---
         
         ## Dependencies
         
         {placeholder -- will be updated in Pass 2 with actual issue numbers}
         {if feature.dependencies: "Depends on: {list dependency feature names}"}
         {if no dependencies: "No dependencies -- can be implemented independently."}
         
         ---
         
         ## Verification Commands
         

         build: {verification.build or "not configured"} tests: {verification.tests or "not configured"} lint: {verification.lint or "not configured"} typecheck: {verification.typecheck or "not configured"} acceptance: {verification.acceptance or "not configured"}

         
         ---
         *Generated by [claude-harness](https://github.com/panayiotism/claude-harness) PRD breakdown*
         *Breakdown ID: {prdMetadata.breakdown}*
         

      3. Create issue via gh issue create or mcp__github__create_issue:

         • Title: {feature.name}
         • Body: rich template from step 2
         • Labels: from step 1
         • Handle failures gracefully (log warning, continue with other features)

      4. Update feature entry in active.json:

         • Set github.issueNumber to created issue number
         • Save updated .claude-harness/features/active.json

      5. Rate limiting: 500ms delay between API calls

15.5. Pass 2 -- Update issues with cross-references (after ALL issues are created):

Now that every feature has an assigned `github.issueNumber`, update each issue that has dependencies or is depended upon:

- For each feature with `dependencies` array OR that appears in another feature's `dependencies`:

  1. **Build "Depends on" section**:
     ```markdown
     **Depends on:**
     - #43 -- Add user authentication
     - #45 -- Create database schema
     ```
     Map each dependency feature ID to its `github.issueNumber` and `name`.

  2. **Build "Blocks" section** (reverse lookup):
     - Find all features whose `dependencies` array includes this feature's ID
     ```markdown
     **Blocks:**
     - #44 -- Build user dashboard
     - #46 -- Add admin panel
     ```

  3. **Replace the Dependencies section** in the issue body:
     - Use `gh issue edit {issueNumber} --body "{updated body}"` or GitHub MCP
     - Replace the placeholder dependency section with the actual cross-referenced version

  4. **Rate limiting**: 500ms delay between API calls

This ensures **bidirectional linking**: if feature-002 depends on feature-001, then:
- feature-001's issue shows "**Blocks:** #44 -- feature-002 name"
- feature-002's issue shows "**Depends on:** #43 -- feature-001 name"

**Error Handling** (applies to both Pass 1 and Pass 2):
- GitHub MCP unavailable -> Log warning, skip issue creation entirely but continue with feature creation
- Permission denied -> Log error for specific feature, continue with others
- API rate limit -> Add 500ms delay between requests
- Network error -> Retry 3x with exponential backoff
- Pass 2 update failure -> Log warning (issues exist but without cross-references), continue

Phase 4: Summary & Next Steps

16. Report completion:

    FEATURES CREATED FROM PRD
    PRD Sections: 5
    Features Extracted: 8
    Created Now: 3
    
    GitHub Issues Created: 3
      #43: Add user authentication (MVP)
      #44: Build user dashboard -> depends on #43
      #45: Create database schema
    Cross-references: 2 issues updated with dependency links
    
    Files:
    - PRD input: .claude-harness/prd/input.md
    - Analysis: .claude-harness/prd/analysis.json
    - Breakdown: .claude-harness/prd/breakdown.json
    
    NEXT STEPS:
    1. Start implementation: /flow feature-001
    2. Or batch process: /flow --autonomous
    3. Review analysis: cat .claude-harness/prd/breakdown.json
    

17. Interactive menu (if user doesn't select all):

    • Use AskUserQuestion with multi-select: true
    • Show pending features from breakdown
    • Allow user to start implementing any features

Command Options

Flags

--no-issues

• Skip GitHub issue creation. Features are still created in active.json.
• Useful when GitHub integration is not configured or not needed.
• By default, issues are always created alongside features.

--analyze-only

• Run PRD analysis without creating features
• Useful for review before committing to features

--auto

• Skip feature review confirmation prompt
• Create all extracted features and GitHub issues automatically

--max-features N

• Limit feature creation to top N features by priority
• Useful for phased rollout

Usage Examples

/claude-harness:prd-breakdown "Detailed PRD markdown here..."           # Inline PRD (creates features + issues)
/claude-harness:prd-breakdown @./docs/prd.md                           # File reference (creates features + issues)
/claude-harness:prd-breakdown --file ./docs/prd.md                     # File flag
/claude-harness:prd-breakdown --url https://github.com/.../issues/42  # GitHub issue as PRD source
/claude-harness:prd-breakdown --analyze-only                           # Analysis only (no features, no issues)
/claude-harness:prd-breakdown --auto                                   # No prompts, full automation
/claude-harness:prd-breakdown --max-features 10                        # Top 10 only
/claude-harness:prd-breakdown @./prd.md --no-issues                   # Features only, skip GitHub issues
/claude-harness:prd-breakdown @./prd.md --auto                        # Full automation: analyze + features + issues

Syntax Variations

Syntax	Behavior
/prd-breakdown "markdown text"	Analyze inline PRD, create features + GitHub issues
/prd-breakdown @path/to/file.md	Read PRD from file, create features + issues
/prd-breakdown --file path/to/file.md	Read PRD from file (--flag syntax)
/prd-breakdown --url https://...	Fetch PRD from GitHub issue
/prd-breakdown @file.md --no-issues	Create features only, skip GitHub issues
/prd-breakdown @file.md --auto	Full automation: analyze, create features AND issues
(no args)	Prompt user for interactive input

Error Handling

Scenario	Action
PRD not provided	Prompt via AskUserQuestion
PRD too large (>100KB)	Warn user, ask to focus section
GitHub fetch fails (no MCP)	Fall back to interactive input
Invalid markdown	Parse as plaintext, still extract
Feature ID collision	Use timestamp suffix for uniqueness
Dependency cycle	Report error, suggest manual ordering
GitHub MCP unavailable	Log warning, skip issue creation entirely but continue with feature creation
Issue creation permission denied	Log error for specific feature, continue with others
Issue creation rate limit	Add 500ms delay between requests, continue
Issue creation network error	Retry 3x with exponential backoff
Pass 2 update failure	Log warning (issues exist but without cross-references), continue

Integration with Other Commands

• With /flow: Each created feature can be implemented via /flow feature-XXX
• With /start: Shows PRD analysis summary from prior sessions
• With memory: Records decomposition patterns to procedural memory for future PRDs