| Skill | Description |
|---|---|
| panel-review-loop | Autonomously loop the review-fix-rereview cycle around panel-review — judge which findings are worth fixing, fix them, re-review, repeat until clean, then report what was fixed, what was left alone (and why), and any goal deviations |
| pr-comment-handler | Read every open review comment on a PR, classify each as fix / push-back / defer-to-Linear, drive the work, then post the right reply on each thread (including Fixed in <sha> replies on push) |
| post-panel-review-comments | Triage findings from a panel-review into PR inline comments and Linear tickets via a two-stage select list (auto-falls back to a top-level comment when GitHub rejects an inline) |
| auto-post-panel-review-comments | Zero-touch twin of post-panel-review-comments — post the legitimate findings straight to the PR (mergeable suggestion when the fix is a drop-in, a +1 reaction when a bot already raised it), routing uncertain / out-of-scope ones to Linear, no prompts |
| auto-review | One-shot pipeline: run panel-review, post the findings via auto-post-panel-review-comments, then approve via approve-pr with a short LGTM — but only when the PR passes a strict approval gate (full coverage, LOW/polish-only, sound approach, not a draft, and more; see the skill) |
| approve-pr | Approve the GitHub PR you're discussing (usually after a panel-review) — auto-detect the PR from your branch and stamp it with a short fun body (LGTM or a silly emoji) when no message is given, or your message verbatim when one is |
| optimize-agents-md | Optimize your AGENTS.md (and CLAUDE.md) files according to best practices. Works with monorepos, too |
| panel-review | Fan a code review out to multiple local CLI agents (codex, claude, opencode) in parallel and amalgamate their findings |
The catena-labs links above point to skills that have moved to catena-labs/dev-skills.
MIT