PRP = PRD + codebase intelligence + validation loop
You give the AI a detailed plan with context and validation commands. The AI implements, tests, and self-corrects until everything passes.
| Command | What it does |
|---|---|
/prp-prd |
Create a PRD with implementation phases |
/prp-plan |
Create an implementation plan |
/prp-implement |
Execute a plan with validation loops |
/prp-ralph |
Autonomous loop until all validations pass |
| Command | What it does |
|---|---|
/prp-issue-investigate |
Analyze a GitHub issue, create a plan |
/prp-issue-fix |
Implement the fix |
/prp-debug |
Deep root cause analysis (5 Whys) |
| Command | What it does |
|---|---|
/prp-commit |
Smart commit with natural language targeting |
/prp-pr |
Create a pull request |
/prp-review |
Review a pull request |
/prp-prd "user authentication system"
↓
Creates PRD with phases (stored in .claude/PRPs/prds/)
↓
/prp-plan .claude/PRPs/prds/user-auth.prd.md
↓
Creates implementation plan for next phase
↓
/prp-implement .claude/PRPs/plans/user-auth-phase-1.plan.md
↓
Executes plan, runs validations, fixes failures
↓
Repeat /prp-plan for next phase
Skip the PRD. Go straight to a plan:
/prp-plan "add pagination to the API"
↓
/prp-implement .claude/PRPs/plans/add-pagination.plan.md
/prp-issue-investigate 123
↓
/prp-issue-fix 123
/prp-debug "TypeError: Cannot read property 'x' of undefined"
↓
Creates RCA report with root cause and fix specification
Instead of /prp-implement, use /prp-ralph for fully autonomous execution:
/prp-ralph .claude/PRPs/plans/my-feature.plan.md --max-iterations 20
This runs in a loop:
- Implements the plan
- Runs all validations
- If something fails → fixes it → re-validates
- Keeps going until everything passes
- Exits when done
Go make coffee. Come back to working code (or a progress log).
Cancel with: /prp-ralph-cancel
After implementation:
/prp-commit # Stage and commit with smart message
/prp-pr # Create pull request
/prp-review 123 # Review someone else's PR
.claude/PRPs/
├── prds/ # PRD documents
├── plans/ # Implementation plans
│ └── completed/ # Archived plans
├── reports/ # Implementation reports
├── issues/ # Issue investigations
└── reviews/ # PR reviews
/prp-prd "I want users to be able to like posts"This asks you clarifying questions, does research, and creates a structured PRD with phases.
/prp-plan "add a like button to posts with real-time count updates"Creates a detailed implementation plan with tasks and validation commands.
/prp-ralph .claude/PRPs/plans/like-button.plan.md --max-iterations 15Autonomous execution until done.
/prp-issue-investigate 456
/prp-issue-fix 456/prp-commit typescript files except tests
/prp-pr- Context is king - The more context in your plan, the better the output
- Validation matters - Plans with test commands work better than plans without
- Use Ralph for big stuff - Let it iterate instead of babysitting
- Max iterations - Always set
--max-iterationson Ralph loops - Start specific - "Add OAuth2 with Google" beats "add authentication"
Previous commands like /prp-base-create, /prp-spec-create, /api-contract-define, etc. are preserved in old-prp-commands/ for reference. The new streamlined flow replaces all of them.
- Big feature? →
/prp-prd→/prp-plan→/prp-ralph - Medium feature? →
/prp-plan→/prp-implement - GitHub issue? →
/prp-issue-investigate→/prp-issue-fix - Weird bug? →
/prp-debug "error message" - Done? →
/prp-commit→/prp-pr
Happy building.