🔓 unclaimed — this page was auto-generated from GitHub. Are you the creator?
Claim this page →
loop-skills
Autonomous, self-re-prompting agent loops for Claude Code — packaged as independently-installable plugins (spec-development, dev-team, pr-autopilot, triangul…
git clone https://github.com/donnfelker/loop-skills
loop-skills
Autonomous, self-re-prompting agent loops for Claude Code. Each plugin here is a bounded autonomous loop that drives an artifact to a terminal state — it iterates, checks a termination condition (a quiet round, two clean passes, a cap), keeps itself alive, and commits per round for an audit trail.
Built by Donn Felker.
Contributions welcome! Found a way to improve a plugin or have a new one to add? Open a PR.
Run into a problem or have a question? Open an issue.
What are Plugins?
Plugins are packages of skills, commands, agents, and hooks that extend Claude Code with specialized capabilities. Each plugin in this marketplace is installed independently — pick only what you need. A few of them compose: spec-development orchestrates the standalone dev-team and pr-autopilot loops (see the dependency note below).
Available Plugins
| Plugin | Description |
|---|---|
| spec-development | End-to-end spec workflow. Bundles plan-to-tickets (import a structured plan into ClickUp, Linear, Jira, Asana, Notion, GitHub Projects, or a markdown fallback as a ticket hierarchy) and implement-full-spec (turn a parent ticket with N actionable subtasks into N stacked PRs and drive each to merge-ready through multi-round bot and human review). Depends on the dev-team and pr-autopilot plugins — install all three together. |
| dev-team | Drive a single unit of work — one spec, ticket, finding, or change request — from spec to a committed, reviewed result with a dev team of coordinating agents: Dev → QA → Reviewer in a bounded cycle. Runs as a real agent team where the harness supports one, else coordinated subagents. Usable standalone, or as the per-subtask engine behind implement-full-spec |
| pr-autopilot | Autonomously watch a single PR and loop through review rounds — fetch every unaddressed review/comment, fix the actionable ones, commit, push, reply/resolve threads, re-request review — until a quiet round or a safety cap. Includes a --single-pass mode (one autonomous round, no loop) for a one-shot comment sweep. Owns the shared pr-review-mechanics.md reference |
| triangulated-code-review | Triangulated multi-reviewer code review orchestrator. Borrows from research methodology — checks each finding against multiple independent reviewers (comprehensive, security, codex, codex adversarial) to reduce blind spots, then runs a QA analyst pass that substantiates every finding (verifying any third-party library claims via the context7 MCP) and demotes unsubstantiated ones into a dedicated "Invalidated Findings" section of the prioritized, timestamped report |
| multi-llm-convergence | Drives any artifact (plan, PRD, design doc, spec, or implementation diff) to convergence by alternating two genuinely different LLM reviewers — a Codex reviewer and an independent Claude review subagent — applying each round's findings and looping until both independently agree it clears the bar. Grounds offline reviewers in local source-of-truth clones, runs a liveness watchdog so a silent reviewer can't hang the loop, and commits per round for an auditable trail |
| multi-llm-convergence-beta | (beta/experimental) Direct-CLI, configurable, N-model variant of multi-llm-convergence. Reaches every model through its own CLI (claude, codex, gemini, grok) instead of the Codex companion script, driven by a JSON adapter registry you extend with a data edit. The operator picks which models converge (≥2 distinct families) and the host is a pure orchestrator. Keeps the original's guarantees — local grounding, a liveness watchdog, per-round commits, and a hard preflight stop (reachable + read-only proven by a negative probe). Ships parallel to multi-llm-convergence (untouched) until promotion |
Installation
Option 1: CLI Install (Recommended)
Use npx skills to install skills directly:
# Install everything
npx skills add donnfelker/loop-skills
# Install specific skills
npx skills add donnfelker/loop-skills --skill dev-team pr-autopilot
# List available skills
npx skills add donnfelker/loop-skills --list
This installs to your .agents/skills/ directory (and symlinks into .claude/skills/ for Claude Code compatibility).
Option 2: Claude Code Plugin
Install via Claude Code's built-in plugin system:
# Add the marketplace
/plugin marketplace add donnfelker/loop-skills
# Install the plugins you want
/plugin install dev-team@loop-skills
/plugin install pr-autopilot@loop-skills
/plugin install spec-development@loop-skills
/plugin install triangulated-code-review@loop-skills
/plugin install multi-llm-convergence@loop-skills
Each /skill still works standalone once installed — /dev-team, /pr-autopilot, /implement-full-spec, /plan-to-tickets, /triangulated-code-review, /multi-llm-convergence.
Plugin dependencies
spec-development's implement-full-spec orchestrates two skills that ship as separate plugins:
- It runs
dev-teamonce per subtask (the Dev → QA → Reviewer → commit loop). - It delegates each PR's review-response mechanics to
pr-autopilot(--single-pass).
Claude Code plugins don't auto-install dependencies, so install dev-team and pr-autopilot alongside spec-development. implement-full-spec checks for both at the start of a run and tells you what to install if either is missing.
License
MIT — see LICENSE.
Contributing
See CONTRIBUTING.md and AGENTS.md for skill-authoring conventions and the PR checklist. In short: name must match the skill's directory, description is a single line under 1024 chars with trigger phrases, SKILL.md stays under 500 lines, and every change updates README.md (the table above) and CHANGELOG.md.
// compatibility
| Platforms | cli |
|---|---|
| Operating systems | — |
| AI compatibility | claude |
| License | MIT |
| Pricing | open-source |
| Language | Shell |
// faq
What is loop-skills?
Autonomous, self-re-prompting agent loops for Claude Code — packaged as independently-installable plugins (spec-development, dev-team, pr-autopilot, triangulated-code-review, multi-llm-convergence).. It is open-source on GitHub.
Is loop-skills free to use?
loop-skills is open-source under the MIT license, so it is free to use.
What category does loop-skills belong to?
loop-skills is listed under plugins in the Claudeers registry of Claude-compatible tools.
// embed badge
[](https://claudeers.com/loop-skills)
// retro hit counter
[](https://claudeers.com/loop-skills)
// reviews
// guestbook
// related in Claude Plugins
🔓A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
🔓Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explainin…
🔓"CLI-Anything: Making ALL Software Agent-Native" -- CLI-Hub: https://clianything.cc/
🔓A Claude Code plugin that shows what's happening - context usage, active tools, running agents, and todo progress