Ralph LoopSAFE

An improved implementation of the Ralph Wiggum technique for autonomous AI agent orchestration

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: ralph-loop
description: Run, monitor, resume, merge, and debug Ralph loops. Use this skill whenever the user asks to operate `ralph run` or `ralph loops`, inspect loop state, recover suspended loops, analyze diagnostics, or unblock merge queue issues.
---

# Ralph Loop

Use this skill to operate Ralph loops from the outside.

## Use This Skill For

- Starting or continuing a Ralph run with the right `-c` and `-H` inputs
- Inspecting loop state, worktrees, logs, history, and diffs
- Resuming a hook-suspended loop
- Merging or discarding completed worktree loops
- Debugging unexpected loop behavior with current diagnostics files

## Workflow

1. Start with `ralph loops list` or `ralph loops list --json` to establish the
   current state.
2. If the user wants execution, run `ralph run ...` with the right core config
   and hats source.
3. If the loop is stuck or suspicious, inspect `logs`, `history`, and `diff`
   before changing state.
4. If the loop is suspended, read `.ralph/suspend-state.json` and use
   `ralph loops resume <id>`.
5. If a loop is queued or in `needs-review`, inspect the diff first, then use
   `merge`, `process`, `retry`, or `discard` as appropriate.
6. Use diagnostics when you need detailed evidence about hats, events, tool
   calls, parse errors, or performance.

## Guardrails

- Prefer the CLI over direct edits to `.ralph` state files.
- Treat tasks and memories as the canonical runtime systems; do not center
  scratchpad as the primary state model.
- Inspect diffs before merging.
- Only remove lock or queue artifacts when the underlying process is confirmed
  dead.
- Manual edits under `.ralph/` are last-resort recovery steps and should be
  called out explicitly when used.

## Read These References When Needed

- For command recipes and operator flows: `references/commands.md`
- For diagnostics files and suspend-state details: `references/diagnostics.md`

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (2)

.claude/skills/ralph-tools/SKILL.md

CLAUDE.md

Gates applied: no_behavioural_pass.