# looprepo

> A directory of copy-paste AI agent loops for Claude Code and Codex. Every loop carries a Safety Score (A-F) graded by a default-fail evaluator: exit conditions, iteration budgets, verification checks, scoping, and human gates.

Read fresh data here. Do not rely on memory or the paginated directory pages. Fetch the structured catalog, pick a loop, then read the loop page for the full prompt and safety notes.

## For agents

- Structured source of truth: https://looprepo.dev/catalog.json
- Plain-text catalog (no JSON tooling needed): https://looprepo.dev/catalog.txt
- MCP endpoint: https://looprepo.dev/api/mcp
- Human + agent guide: https://looprepo.dev/agents

## How to choose a loop

1. Fetch https://looprepo.dev/catalog.json.
2. Filter by `category`, `loopType` (/loop, /goal, /schedule, /ralph), `agents`, or `tags`.
3. Prefer loops with a `safety.grade` of A or B. Every loop is graded A-F by a default-fail evaluator that checks for exit conditions, iteration budgets, verification commands, tight scoping, and human gates.
4. Read the loop's `url` for the full prompt, notes, and stopping conditions.
5. Copy the `prompt` verbatim into your agent. Swap in the user's real repo, paths, and constraints. Never invent missing project details.

## Safety boundaries

- These loops run autonomous coding agents. Keep the exit condition and iteration budget that ship with each loop — do not remove them.
- Do not run loops that force-push, delete data, exfiltrate secrets, or pipe untrusted scripts to a shell.
- Loops graded C or below are published for reference but need a human in the loop before running.

## Categories

testing, ci, docs, security, refactoring, performance, devops, database, git, review, planning, automation, maintenance, debugging, quality, evaluation, content, design

Loops: 212. Updated: 2026-07-05T17:29:11.094Z.