Blueprint Prp Create

/blueprint:prp-create

Create a comprehensive PRP (Product Requirement Prompt) - a self-contained packet with all context an AI agent needs to deliver production code on first attempt.

What is a PRP? PRD + Curated Codebase Intelligence + Implementation Blueprint + Validation Gates = everything needed for reliable implementation.

Usage: /blueprint:prp-create [feature-name]

Prerequisites:

• Blueprint Development initialized (docs/blueprint/ exists)
• Clear understanding of the feature to implement

When to Use This Skill

Context

• Blueprint initialized: !find . -path '*/docs/blueprint/*' -maxdepth 3 -name 'manifest.json' -type f
• Last PRP ID: !find . -path '*/docs/blueprint/manifest.json' -maxdepth 3 -exec jq -r '.id_registry.last_prp // 0' {} +
• Curated rules available: !find . -maxdepth 3 -path '*/.claude/rules/*' -type f -name "*.md"
• Existing PRDs: !find . -path '*/docs/prds/*' -name "*.md" -type f

Parameters

Parse $ARGUMENTS:

• feature-name (required): Kebab-case name for PRP (e.g., auth-oauth2, api-rate-limiting)
  • Used for filename and document ID generation

Execution

Execute the complete PRP creation workflow:

Step 1: Verify prerequisites and understand requirements

1. If Blueprint not initialized → Error: "Run /blueprint:init first"
2. Ask user to describe the feature: "What feature needs to be implemented?"
   • Capture: Goal, why it matters, success criteria
3. Ask if this implements an existing PRD: "Does this implement an existing PRD or is it standalone?"
   • If PRD chosen → Store PRD ID for linking
4. Determine feature type (API endpoint, database, UI, integration, etc.)

Step 2: Research codebase patterns

Use Explore agent to find existing patterns:

1. Similar features already implemented (identify existing patterns to follow)
2. Relevant file locations and integration points
3. Testing patterns used for similar features
4. Architecture decisions that affect this feature

Store findings with specific file paths and line numbers.

Step 3: Research external documentation

For relevant libraries/frameworks, gather:

1. Official documentation sections (capture URLs with specific sections)
2. Known issues and gotchas from Stack Overflow / GitHub discussions
3. Best practices from documentation
4. Common implementation patterns

Use WebSearch/WebFetch as needed. Create or update curated rule entries if needed (see REFERENCE.md).

Step 4: Generate PRP document ID and structure

Generate next PRP ID from manifest:

• Extract id_registry.last_prp from manifest.json
• Next ID = last_prp + 1 (format: PRP-NNN)

Create docs/prps/[feature-name].md with frontmatter and sections (see REFERENCE.md).

Step 5: Draft PRP content with research findings

Fill all required sections (see REFERENCE.md):

1. Goal & Why: One-sentence goal, business justification, target users, priority
2. Success Criteria: Specific, testable acceptance criteria with metrics
3. Context:
   • Documentation references (URLs with specific sections)
   • Curated rule references (links to .claude/rules/ library/pattern entries)
   • Codebase Intelligence: File paths, code snippets with line numbers, patterns to follow
   • Known Gotchas: Critical warnings with mitigations
4. Implementation Blueprint:
   • Architecture decision with rationale
   • Task breakdown (Required / Deferred / Nice-to-Have categories)
   • Order of implementation
5. TDD Requirements: Test strategy and critical test cases
6. Validation Gates: Executable commands (linting, type-checking, tests, coverage)

Critical: All tasks must be explicitly categorized (see REFERENCE.md).

Step 6: Score confidence across dimensions

Rate each dimension 1-10:

Calculate overall score as average of dimensions. Target: 7+ for execution, 9+ for subagent delegation.

If score < 7 → Return to Steps 2-3 to fill gaps.

Step 7: Review and validate completeness

Verify checklist (see REFERENCE.md):

• Goal is clear and specific
• Success criteria are testable
• All file paths are explicit (not "somewhere in...")
• Code snippets show actual patterns with line references
• Gotchas include mitigations
• Validation commands are copy-pasteable
• Confidence score is honest

Update docs/blueprint/manifest.json ID registry with new PRP entry.

Step 8: Report PRP and prompt for next action

Display summary showing:

• PRP ID and location
• Feature summary and approach
• Context collected (curated rules, patterns, documentation)
• Linked documents (source PRD if applicable)
• Confidence score with breakdown
• Any gaps if score < 7

If confidence >= 7, offer user choices:

• Execute PRP now → /blueprint:prp-execute [feature-name]
• Create work-order for subagent → /blueprint:work-order
• Review and refine → Show file location and gaps
• Done for now → Exit (save for later execution)

If confidence < 7, offer user choices:

• Research more context → Use Explore agent for gaps
• Create curated rule entries → /blueprint:curate-docs
• Execute anyway (risky) → Proceed with warning
• Done for now → Save incomplete PRP

Agentic Optimizations

For PRP document structure, task categorization, review checklists, and curated-rule creation guidance, see REFERENCE.md.