/blueprint-prp-create

Create a comprehensive PRP (Product Requirement Prompt) for a feature or component.

What is a PRP? A PRP is PRD + Curated Codebase Intelligence + Implementation Blueprint + Validation Gates - the minimum viable packet an AI agent needs to deliver production code successfully on first attempt.

Prerequisites:

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

Steps:

Phase 1: Research

1.1 Understand Requirements

Ask the user or analyze context to understand:

• Goal: What needs to be accomplished?
• Why: What problem does this solve?
• Success Criteria: How do we know it's done?

1.2 Research Codebase

Explore the existing codebase to understand:

• Existing patterns: How similar features are implemented
• Integration points: Where this feature connects
• Testing patterns: How similar features are tested
• File locations: Where new code should go

Use the Explore agent:

<Task subagent_type="Explore" prompt="Research patterns for [feature type] implementation">

1.3 Research External Documentation

For any libraries/frameworks involved:

• Search for relevant documentation sections
• Look for known issues and gotchas
• Find best practices and patterns

Use WebSearch/WebFetch to gather:

• Official documentation for key libraries
• Stack Overflow discussions about common issues
• GitHub issues for known problems

1.4 Check/Create ai_docs

Look for existing ai_docs:

ls docs/blueprint/ai_docs/libraries/
ls docs/blueprint/ai_docs/project/

If relevant ai_docs don't exist, create them:

• Extract key patterns from documentation
• Document gotchas discovered in research
• Create curated, concise entries (< 200 lines)

Phase 2: Draft PRP

2.1 Create PRP File

Create the PRP in docs/prps/:

docs/prps/[feature-name].md

2.2 Fill Sections

Goal & Why:

• One sentence goal
• Business justification
• Target users
• Priority

Success Criteria:

• Specific, testable acceptance criteria
• Performance baselines with metrics
• Security requirements

Context:

• Documentation References: URLs with specific sections
• ai_docs References: Links to curated docs
• Codebase Intelligence:
  • Relevant files with line numbers
  • Code snippets showing patterns to follow
  • Integration points
• Known Gotchas: Critical warnings with mitigations

Implementation Blueprint:

• Architecture decision with rationale
• Task breakdown with pseudocode
• Implementation order

TDD Requirements:

• Test strategy (unit, integration, e2e)
• Critical test cases with code templates

Validation Gates:

• Executable commands for each quality gate
• Expected outcomes

Phase 3: Assess Confidence

3.1 Score Each Dimension

Dimension	Scoring Criteria
Context Completeness	10: All file paths, snippets explicit. 7: Most provided. 4: Significant gaps
Implementation Clarity	10: Pseudocode covers all cases. 7: Main path clear. 4: High-level only
Gotchas Documented	10: All known pitfalls. 7: Major gotchas. 4: Some mentioned
Validation Coverage	10: All gates have commands. 7: Main commands. 4: Incomplete

3.2 Calculate Overall Score

• Average of all dimensions
• Target: 7+ for execution, 9+ for subagent delegation

3.3 If Score < 7

• Research missing context
• Add ai_docs entries
• Document more gotchas
• Add validation commands
• Clarify pseudocode

Phase 4: Review

4.1 Self-Review Checklist

• 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 executable
• Confidence score is honest

4.2 Present to User

Show the user:

• PRP summary
• Key implementation approach
• Confidence score
• Any areas needing clarification

Output Template:

## PRP Created: [Feature Name]

**Location:** `docs/prps/[feature-name].md`

**Summary:**
[1-2 sentence summary of what will be implemented]

**Approach:**
- [Key architectural decision]
- [Main implementation pattern]

**Context Collected:**
- [X] ai_docs entries: [list]
- [X] Codebase patterns identified
- [X] External documentation referenced
- [X] Known gotchas documented

**Validation Gates:**
- Gate 1: [Linting command]
- Gate 2: [Type checking command]
- Gate 3: [Unit tests command]
- Gate 4: [Integration tests command]

**Confidence Score:** X/10
- Context: X/10
- Implementation: X/10
- Gotchas: X/10
- Validation: X/10

**Needs attention (if score < 7):**
- [List any gaps to address]

4.3 Prompt for next action (use AskUserQuestion):

If confidence score >= 7:

question: "PRP ready (confidence: X/10). What would you like to do?"
options:
  - label: "Execute PRP now (Recommended)"
    description: "Implement the feature with TDD workflow and validation gates"
  - label: "Create work-order for subagent"
    description: "Package this PRP for isolated execution by a subagent"
  - label: "Review and refine first"
    description: "I want to improve the PRP before executing"
  - label: "I'm done for now"
    description: "Save PRP and exit - execute later with /blueprint:prp-execute"

If confidence score < 7:

question: "PRP needs work (confidence: X/10). What would you like to do?"
options:
  - label: "Research more context"
    description: "Explore codebase and documentation to fill gaps"
  - label: "Create ai_docs entries"
    description: "Curate library documentation to improve context"
  - label: "Execute anyway (risky)"
    description: "Proceed with implementation despite low confidence"
  - label: "I'm done for now"
    description: "Save incomplete PRP and return later"

Based on selection:

• "Execute PRP now" → Run /blueprint:prp-execute [feature-name]
• "Create work-order" → Run /blueprint:work-order
• "Review and refine" → Show PRP file location and key gaps
• "Research more context" → Use Explore agent on identified gaps
• "Create ai_docs entries" → Run /blueprint:curate-docs for relevant libraries
• "Execute anyway" → Run /blueprint:prp-execute [feature-name] with warning
• "I'm done" → Exit

Tips:

• Be thorough in research phase - it saves implementation time
• Include code snippets with actual line numbers
• Document gotchas as you discover them
• Validation gates should be copy-pasteable commands
• Honest confidence scoring helps decide next steps

Error Handling:

• If docs/blueprint/ doesn't exist → Run /blueprint:init first
• If libraries unfamiliar → Research documentation thoroughly
• If codebase patterns unclear → Use Explore agent extensively