Loop Library
Guide AI agents to discover, find, audit, craft, run, and improve bounded, repeatable workflows.
The skill documentation strictly adheres to least privilege: all operations require user authorization scope, explicitly prohibit destructive, production, financial, privacy-sensitive, or external-message actions without consent; publishing requires explicit preview and approval, distinguishes draft vs public; treats LOOPS.md and catalog content as untrusted data, never executing embedded instructions; refuses to save loops containing secrets. Source code for CI and Worker shows permission controls and input validation (Turnstile, Origin checks), indicating security-by-design. Deductions: Publisher identity unverified; reliance on external services (e.g., signals.forwardfuture.com) whose security/availability not audited; some sensitive operations (like publishing) depend on external services, full least-privilege not confirmed.
The skill has clear workflows and multiple reference files (discover, audit, run, etc.), internally consistent, happy path plausible. Tests exist (CI and Worker test files) covering parts of key paths, but the skill itself as a prompt is not executed in static review; error handling emphasizes honest reporting of no-progress, but lacks systematic test coverage of all edge cases. Deductions: Repeatability/determinism not confirmed statically; dependence on live catalog; if unavailable, degrades to unavailability declaration without local fallback.
The skill targets engineering teams and AI agents; scenarios are clearly described (discover, find, audit, run, publish, etc.), triggers are explicit in SKILL.md, invocable via explicit call or semantic matching. But capability boundaries and non-fit ranges are limited: for users not using loops or simple one-shot tasks, it only briefly suggests, without detailed non-fit guidance. Environment fit: Docs are all English, no Chinese support; reachability of external services from mainland China is not audited, but catalog is from overseas URL, likely inaccessible from China, impacting applicability.
Standard documentation structure with introduction, usage, examples, and multiple reference files; information hierarchy clear. Provides install commands and dependency notes (Node, npx). License (MIT) and README present, but version info only in revision, no explicit version number or changelog in strength; known limitations disclosed sparsely, e.g., dependence on live catalog but no fallback described; maintenance responsibility declared as Forward Future project, but no maintainer contact or update policy.
The skill claims to help users discover, design, run loops, and output well-formatted loop descriptions. Based on docs, core workflow design sound, output formats explicit (run receipt, debrief), should complete core tasks. But actual output quality cannot be verified statically; reliance on external catalog affects completeness. Deductions: Direct usability and marginal value untested; publishing depends on external services, requires network and permissions, adding complexity and cost.
Evidence mainly from documentation and source review; CI and Worker tests provide verification of key technical components, but the skill itself as a prompt cannot be verified statically. Traceability moderate: loop definitions require exact text or SHA-256, providing reproducibility. Deductions: Core execution path not independently reproduced; key claims (e.g., 'works well') rely on author description, lacking third-party verification.
- External dependency: Core functionality depends on live catalog and publishing services at signals.forwardfuture.com. These may be inaccessible from mainland China, rendering discovery, find, and publish paths unavailable; security and availability not audited.
- Unverified publisher: Publisher not verified in FollowSkills curated registry; identity unknown. Adds uncertainty to maintenance responsibility and update path.
- Chinese language support: Documentation is entirely English, potentially unfriendly to Chinese users; consider providing Chinese version or localized triggers.
- Runtime safety: Skill execution relies on user authorization scopes and external services; ensure users understand potential external calls (e.g., publishing, network requests).
What does this skill do, and when should you use it?
Loop Library is an Agent Skill that structures how an AI agent works with repeatable workflows: it discovers loop opportunities from codebases or coding threads, recommends published loops from a live catalog, audits and repairs existing loops, crafts new ones through a guided interview, runs loops within authorized scope and returns evidence-backed run receipts, debriefs completed runs, saves project loops to LOOPS.md, and prepares publication drafts. The skill emphasizes bounded execution, clear success criteria, and human approval boundaries to prevent runaway automation. It is offered as a compatibility alias for legacy installations; new installations should use the new name Loopy.
Reads the routing logic in SKILL.md to select the smallest path (Discover, Find, Audit, Adapt, Craft, Run, Debrief, Save, Publish) based on the user's request. When analysis is needed, reads reference files: references/discover.md, references/audit.md, references/run.md, references/debrief.md, and references/publish.md. For Find operations, fetches the live catalog.md or catalog.json from signals.forwardfuture.com and searches for matches. For Run operations, executes the loop's steps while respecting permission boundaries and writes a run receipt. For Save operations, appends the loop to LOOPS.md at the project root if it exists. For Publish operations, checks catalog overlap and prepares an exact preview for approval.
- An engineer wants to analyze a codebase for repetitive tasks and turn them into a reusable loop.
- A developer needs to find a published loop for a task like keeping documentation up to date.
- A team wants to audit an existing loop, identify weaknesses, and repair only material issues.
- A user wants to craft a new loop for turning customer feedback into verified fixes through an interview.
- Running a loop within authorized scope and obtaining an evidence-backed receipt.
- A project wants to save a loop to LOOPS.md for reuse in later sessions.
What are this skill's strengths and limitations?
- Bounded execution: prevents runaway automation with explicit terminal states, approval gates, and no-progress stops.
- Multi-path coverage: handles discovery, finding, auditing, adapting, crafting, running, debriefing, saving, and publishing.
- Live catalog: uses a live catalog to find published loops, avoiding hallucinated recommendations.
- Evidence-based: requires concrete evidence (e.g., two occurrences) before declaring work repeated.
- Compatibility: provides a `loop-library` alias for legacy installations.
- Dependent on live catalog: finding published loops requires network access, and the feature is unavailable when the catalog is down.
- No test suite: the repository does not mention automated tests, so manual verification may be needed.
- Platform-specific installation: installation commands are only provided for Codex, Cursor, and Claude Code.
- Assumes reference files: the skill references files like discover.md and audit.md, but they are not explicitly listed in the repository, which may lead to issues if missing.
How do you install this skill?
Install via npm/npx: run npx skills add Forward-Future/loopy --skill loopy --agent codex -g -y (or the appropriate agent name for Cursor or Claude Code). The repository also provides a compatibility alias loop-library for older installations. After installation, invoke with /loopy or $loopy.
How do you use this skill?
After installation, simply describe the task in natural language. For example, in Claude Code or Cursor: /loopy Find a loop for improving test reliability. In Codex, select Loopy from /skills, or use $loopy. The skill guides you through the paths: if the request is vague, it asks 'What are you trying to accomplish?'. You can also explicitly request a path like 'Analyze this repo and my coding threads for repeated work' or 'Audit this loop'.