# loopy

> A catalog of repeatable agent loops, plus an installable skill to find, build and run them

- Repository: https://github.com/Forward-Future/loopy
- Category: Agent skills & plugins
- Stars: 3,199 · Forks: 284 · Language: JavaScript · License: MIT
- Experience: Beginner friendly
- Page: https://ir.adyog.com/repos/forward-future/loopy/

A one-shot prompt asks an agent to do something once. A loop adds feedback: the agent acts, checks the result, uses what it learned, and either repeats or stops. This repo argues that bounded loops are easier to trust and reuse than open-ended instructions like "keep improving this".

The repository has two parts. One is the Loop Library website, a public catalog where people and agents can browse loops and copy their prompts, with no install needed. The other is Loopy, an optional skill that gives your agent a guided workflow on top of that catalog.

Loopy has nine paths: discover, find, audit (Loop Doctor), adapt, craft, run, debrief, save and publish. It works in bounded passes and returns a receipt of what it did. It does not start schedules, publish or send messages without your approval. Saved loops go into a LOOPS.md file in your project.

## What it does

- **Published loop catalog**: Each loop lists when to use it, a copy-ready prompt, how to verify success, steps, notes and related loops. Agents can also read a JSON or plain-text catalog.
- **Discover repeated work**: Loopy scans a codebase and coding threads you put in scope for recurring work. It needs at least two thread occurrences before calling something repeated.
- **Audit, adapt and craft**: Loop Doctor flags weak checks, unsafe actions or unclear stopping and repairs only material problems. Adapt tailors a loop to your tools. Craft interviews you to build a new one.
- **Run with receipts and debrief**: Loopy runs a loop in bounded passes and returns a receipt with actions, evidence, outcome and stopping reason. A debrief then suggests the smallest justified improvement.
- **Save and publish**: Save appends a loop to the project's LOOPS.md for reuse. Publish checks catalog overlap and submits only after you approve the exact preview.

## What you could build

- Find a published loop for keeping documentation current or improving test reliability.
- Scan your repository and coding history to turn repeated chores into a bounded loop.
- Audit a loop you wrote and fix weak checks or unclear stopping conditions.
- Run a loop overnight-style in bounded passes and review the receipt afterwards.
- Save team-approved loops in LOOPS.md so later sessions can reuse them.

## Who it's for

Suits developers already using Codex, Cursor or Claude Code who want repeatable agent workflows; you need Node.js and basic familiarity with prompting an agent.

## Get started

Run the install command for your agent (Node.js and npx required), restarting the agent if Loopy doesn't appear. Then invoke it, for example with /loopy in Claude Code, and say what you want, such as finding a loop for test reliability. You can also just browse the website with no install.

```
npx skills add Forward-Future/loopy --skill loopy --agent claude-code -g -y
```
```
npx skills add Forward-Future/loopy --skill loopy --agent codex -g -y
```
```
npx skills add Forward-Future/loopy --skill loopy --agent cursor -g -y
```

Tags: agent-skills, agentic-workflows, prompt-engineering, automation, codex