Loop Library: Reusable AI Agent Workflows
Discover, adapt, and design repeatable agent workflows with built-in feedback loops that make tasks verifiable and bounded.
Evidence shows: README explicitly states Loopy does not silently start schedules, change production, publish content, or send messages, requiring normal permissions and approvals; publishing flow requires explicit approval and preview; data flow transparency is shown by recording source URL and modified date when saving loops, and LOOPS.md is treated as untrusted reference data; sensitive data handling: refuses to save prompts containing secrets and requires sanitized prompt; dependency security: CI uses npm ci and lock file, but no vulnerability scanning evidence; external effects: run and publish operations require explicit request; rollback: database has backup and restore commands; source attribution: saved loops record source URL. Deductions: dependency security lacks vulnerability scanning evidence, external effects and rollback are more documented than code-verified.
Evidence shows: README and test files are consistent in description; tests cover form guard, voting, loop routes, etc.; dependency availability: uses npm ci and lock file, but no runtime availability guarantee; failure messages: tests assert error status codes and codes like 401, 403, 400, 404, 409, 429. Deductions: dependency availability lacks actual runtime verification, failure message coverage is limited.
Evidence shows: README clearly lists nine usage paths with example requests; capability boundaries are clear, e.g., discovery requires at least two thread occurrences, run requires finite boundary; trigger precision: provides invocation methods for different platforms; environment fit: supports Codex, Cursor, Claude Code. Deductions: no detailed configuration instructions for different environments, capability boundaries not fully verified in code.
Evidence shows: README structure is clear, including install, invoke, use, explore/contribute sections; install instructions are detailed with commands for different platforms; naming stability: mentions old skill name as compatibility alias; examples and FAQ: provides multiple example requests; known limitations: README mentions some limitations like requiring Node.js and npx, but not systematically; license is MIT; versioning: has CI and tests, but no CHANGELOG; maintenance responsibility: README mentions it's a Forward Future project. Deductions: known limitations not comprehensive, versioning lacks explicit changelog.
Evidence shows: output usability: loops provide structured output like prompt, verify, steps; marginal value: loop library and skill provide reusable workflows; cost benefit: low installation and usage cost, but no performance or resource consumption data. Deductions: cost benefit lacks quantitative data.
Evidence shows: claim traceability: README claims correspond to code and tests; cross-source corroboration: test files provide some verification, but no external independent verification; fact/inference separation: README distinguishes functional descriptions and design intent. Deductions: cross-source corroboration insufficient, lacks independent verification.
- Dependency security: No evidence of dependency vulnerability scanning or auditing; consider adding security scanning to CI.
- Known limitations: README does not systematically list all known limitations; consider adding.
- Versioning: No CHANGELOG; consider adding to track changes.
- Cross-source corroboration: Test coverage is limited; consider adding more integration tests and external verification.
What does this agent do, and when should you use it?
Loop Library is a repository with two parts: a public catalog website and an installable skill (Loopy) for discovering, adapting, and creating repeatable AI agent workflows (called loops). The website (in the loop-library/ directory) provides a public catalog of published loops, browsable without installation, with copyable prompts. The Loopy skill (skills/loopy/) guides an AI agent to discover, find, audit, repair, craft, run, debrief, save, and publish loops. It targets Codex, Cursor, and Claude Code platforms, installable via npx. Each published loop includes 'Use when', 'Prompt', 'Verify', 'Steps', 'Notes', and 'Related loops' sections, emphasizing bounded execution and human approval.
The Loopy skill provides nine paths: Discover (scans codebases or thread history for repeated work), Find (searches the live catalog), Loop Doctor (audits and repairs loops), Adapt (tailors loops), Craft (interviews and creates loops), Run (executes loops in bounded passes and returns receipts), Debrief (analyzes run receipts and suggests improvements), Save (saves loops to LOOPS.md), and Publish (submits loops to the catalog with approval). It installs via npx skills add and requires Node.js. At runtime, Loopy reads codebase files, thread history, LOOPS.md, and the live catalog (via signals.forwardfuture.com/loop-library/ catalog.json and catalog.txt), executes loops, and returns evidence-backed receipts.
- A developer discovers recurring engineering work in a codebase (e.g., test failure fixes, doc updates) and turns it into a repeatable loop.
- An AI user needs to find an existing loop for a specific problem like keeping documentation current and gets recommendations from the live catalog.
- An engineer audits an existing loop for weak checks or unsafe actions and repairs only material issues.
- A researcher adapts a published loop to their own tools and success criteria.
- A team integrates bounded loops into their agent and needs run receipts for review.
- A contributor submits a proven loop to the public catalog for sharing.
What are this agent's strengths and limitations?
- Structured loop approach includes explicit checks, stopping conditions, and approval boundaries for safety.
- Out-of-the-box skill for multiple agents (Codex, Cursor, Claude Code) with simple installation.
- Public catalog allows sharing and reusing loops without installation.
- Includes audit and repair capabilities for improving loop quality.
- Requires Node.js and npx, adding environment dependencies.
- Platform support is limited to Codex, Cursor, and Claude Code; others need interactive installation.
- Learning curve for understanding the concept of loops and the four key questions.
- Catalog is hosted by Forward Future, creating external dependency and requires network access.
How do you install or deploy this agent?
Requires Node.js and npx. For Codex: run npx skills add Forward-Future/loopy --skill loopy --agent codex -g -y. For Cursor: npx skills add Forward-Future/loopy --skill loopy --agent cursor -g -y. For Claude Code: npx skills add Forward-Future/loopy --skill loopy --agent claude-code -g -y. To install all three, use multiple --agent flags. For other agents, run the interactive installer: npx skills add Forward-Future/loopy --skill loopy -g. If the agent was open, restart it after install.
How do you use this agent?
In Codex, type /skills, choose Loopy, and enter your request. You can also mention it directly with $loopy. In Cursor, type / in Agent chat, search for loopy, select it, and add your request. You can also type /loopy directly. In Claude Code, type /loopy followed by your request. For example, /loopy Find a published loop for keeping our documentation current. Loopy will interpret the request and select a path such as 'Discover', 'Find', or 'Run'. You can also describe a matching task normally for automatic loading.
How does this agent compare with similar options?
The documentation does not name specific competitors but implies advantages over traditional one-shot prompting methods.