ralph-claude-loop

Loop (Claude Code)

Start a Loop optimized for Claude Code. This skill sets up the loop state, activates the official stop hook, and begins working on the first pending story from prd.json.

Prerequisites

• The loophaus plugin must be installed
• A prd.json file must exist in the project root (generate one with /loop-plan)

How It Works

1. This skill writes .loophaus/state.json (the state file)
2. loophaus's stop hook reads this file and intercepts session exits
3. Each iteration follows the Implement → Discover → Plan → Continue cycle
4. When all stories pass: output <promise>COMPLETE</promise>

Iteration Cycle (per story)

┌─ IMPLEMENT ─────────────────────────────┐
│ Pick next story → implement → verify    │
└─────────────────────┬───────────────────┘
                      ▼
┌─ DISCOVER ──────────────────────────────┐
│ Review what you just built.             │
│ Did you find:                           │
│  - Hidden complexity?                   │
│  - Missing edge cases?                  │
│  - New dependencies?                    │
│  - Broken assumptions from the PRD?     │
└─────────────────────┬───────────────────┘
                      ▼
┌─ UPDATE PRD ────────────────────────────┐
│ If discoveries found:                   │
│  1. Add new stories to prd.json         │
│  2. Log discoveries in progress.txt     │
│  3. Adjust max_iterations if needed     │
│ If nothing new: skip                    │
└─────────────────────┬───────────────────┘
                      ▼
┌─ CONTINUE ──────────────────────────────┐
│ Mark current story done → next story    │
└─────────────────────────────────────────┘

Activation

Run the setup script to create the state file:

mkdir -p .loophaus && cat > .loophaus/state.json << 'LOOP_STATE'
{
  "active": true,
  "prompt": "Read prd.json for the task plan. Read progress.txt for current status (check Codebase Patterns section first). Pick the highest priority user story where passes is false. Implement that ONE story. Run verification: $VERIFY_CMD. On failure: read error, fix, retry up to 3 times. On success: commit with message feat: [Story ID] - [Story Title]. Update prd.json: set passes to true for the completed story. DISCOVERY PHASE — After completing the story, review what you just built: did implementation reveal hidden complexity not covered by existing stories? Are there missing edge cases, error handling, or integration points? Did you discover new dependencies or broken assumptions? Are there follow-up tasks needed for what you just built? If YES to any: add new user stories to prd.json with the next available ID (e.g. US-010), set passes: false, and set priority appropriately. Log the discovery in progress.txt under Discoveries section with rationale. If NO: skip, just append progress to progress.txt with learnings. If ALL stories (including newly added ones) have passes true, output <promise>COMPLETE</promise>. When stuck after 3 retries: set notes field in prd.json with error details, move to next story.",
  "completionPromise": "COMPLETE",
  "maxIterations": $MAX_ITERATIONS,
  "currentIteration": 0,
  "sessionId": ""
}
LOOP_STATE

Replace $MAX_ITERATIONS with the desired limit and $VERIFY_CMD with the project's verification command.

After setup, immediately begin working:

1. Read prd.json
2. Pick the first story with passes: false
3. Implement it — write real code, make real changes
4. Run verification
5. Commit
6. Discovery Phase — review what you just built for hidden complexity, missing cases, new dependencies
7. If discoveries found: add new stories to prd.json, log in progress.txt
8. Update prd.json (mark current story done) and progress.txt
9. The stop hook will re-inject this prompt for the next story

Discovery Phase Rules

• New stories MUST follow the same format (id, title, description, acceptanceCriteria, priority, passes: false)
• Use next sequential ID (if last is US-007, new ones start at US-008)
• Set priority relative to remaining work (urgent discoveries get lower priority number)
• Do NOT add trivial or cosmetic items — only genuinely missing functionality
• Log each discovery in progress.txt: what was found, why it matters, which story revealed it
• If adding stories increases scope significantly, update maxIterations in .loophaus/state.json

max_iterations Auto-Adjustment

When adding N new stories, update maxIterations:

new_max = current_iteration + (remaining_stories + new_stories) * 2 + 3

Only increase, never decrease. Update the value in .loophaus/state.json.

CRITICAL RULE: Only output <promise>COMPLETE</promise> when ALL stories in prd.json (including dynamically added ones) have passes: true. Do not output it prematurely.