loopy: Find and Run Repeatable AI-Agent Workflows

Overview

Loopy is an optional skill for AI coding agents that helps turn recurring work into bounded workflows with feedback, verification, and clear stopping conditions. It can find a suitable published workflow, adapt or create one, run it, and review the result.

The companion Loop Library website is a separate public catalog that can be browsed without installing Loopy. The skill is useful when a one-shot prompt is not enough and you want an agent to check results before taking another step.

Key Features

• Searches the live catalog and recommends published loops without running them.
• Inspects accessible codebases and coding threads for recurring work, then proposes a qualified loop.
• Audits and repairs material problems in a loop, including weak checks or unclear stopping behavior.
• Adapts existing loops or guides a user through crafting a new one.
• Runs workflows in bounded passes and returns a receipt with actions, evidence, outcome, and stopping reason.
• Debriefs run receipts and recommends a minimal, evidence-backed improvement.
• Saves requested project loops in LOOPS.md and can prepare catalog submissions for explicit approval.

Use Cases

• A developer wants to identify repeated maintenance work in a repository and make it reusable as a checked workflow.
• A team wants an agent to improve test reliability or documentation in bounded passes, stopping when checks pass or progress ends.
• An agent user wants to find and adapt a published workflow to their tools and definition of success.
• A workflow author wants to audit an existing prompt for verification, safety, and stopping conditions before sharing it.

Getting Started

Install the skill for Codex, Cursor, or Claude Code with:

npx skills add Forward-Future/loopy --skill loopy --agent codex -g -y

Replace codex with cursor or claude-code for those agents. Installation requires Node.js and npx. See the README for invocation, other installation options, and workflow details. Browse the separate Loop Library without installing the skill.

Alternatives

• Loop Engineering: Unlike loopy's installable skill for finding and running bounded loops, it provides patterns and CLI tools for building agent orchestration systems.

Considerations

• Loopy is a companion skill, not the catalog website; installing it does not install or host the website.
• It can inspect only repositories and coding-thread history available to the agent and placed in scope. If thread history is unavailable, discovery relies on codebase evidence.
• Running loops requires a finite boundary and stops at success, a blocker, an approval boundary, an exhausted limit, or lack of measurable progress.
• Loopy does not independently publish content or take other approval-sensitive actions. Publication and other consequential actions require explicit approval.