ralph-loop-crew

Overview

This skill wraps the ralph-loop plugin with pixl-crew conventions: structured completion criteria, crew agent verification, and context packet output. Instead of raw iteration prompts, it sets up a disciplined loop with measurable exit conditions.

Step 1: Parse Task & Define Completion Criteria

From the user's task description (passed as argument or asked interactively):

1. Extract the task — what needs to be done
2. Define measurable completion criteria — concrete checks that prove the task is done
3. Set iteration budget — default 5, max 15 (ask user if ambiguous)

Examples of completion criteria:

• "fix all lint errors" → npx eslint . --max-warnings 0 exits 0
• "make all tests pass" → npm test exits 0
• "refactor to DDD" → no lint errors + tests pass + reviewer agent finds no issues
• "build the feature" → builds + tests pass + type-check passes

Step 2: Detect Project Tooling

Scan the project for available verification commands:

• package.json → npm test, npm run lint, npm run build, npm run typecheck
• Makefile → make test, make lint, make check
• pyproject.toml → pytest, ruff check, mypy
• Cargo.toml → cargo test, cargo clippy

Build a verification command list from what's available.

Step 3: Write Ralph Loop Configuration

Create .claude/ralph-loop.local.md with the following structure:

---
iteration: 0
max_iterations: { budget }
task: "{task summary}"
---

## Task

{Full task description from user}

## Completion Promise

The loop is complete when ALL of the following pass:

{Numbered list of completion criteria with exact commands}

## Verification Commands

Run these after each iteration to check progress:

```bash
{verification commands, one per line}
```

Instructions

1. Read this file at the start of each iteration to get current state
2. Work on the task — make meaningful progress each iteration
3. Run all verification commands
4. Update the iteration count in frontmatter
5. If all completion criteria pass, announce completion
6. If not, note what remains and continue next iteration