skills-review

Review, approve, or reject staged skill proposals pending in the skills-pending directory.

6|1|Updated Jun 6, 2026
One-click install
npx skills add https://github.com/t-mercier/ai-agents-orchestrator --skill skills-review-t-mercier
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: skills-review
Source: https://github.com/t-mercier/ai-agents-orchestrator/tree/main/skills/skills-review
Command: npx skills add https://github.com/t-mercier/ai-agents-orchestrator --skill skills-review-t-mercier

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Skill proposals made when nobody was present to answer — headless runs, background sessions — sit inert in ~/.claude/skills-pending/ and never take effect. This Skill provides the approval gate: it lists what is waiting, shows the exact content or diff, and promotes or rejects it only after explicit confirmation. ## Core Features & Use Cases - Pending proposal triage: Lists new skills, patches, and archive requests staged in ~/.claude/skills-pending/ with their rationale and proposal dates. - Safe approval workflow: Shows the full SKILL.md or a real diff via patch_apply.py before asking for confirmation, then promotes, applies, or archives atomically. - Learning loop health check: Reports when the loop last produced or applied anything, and diagnoses silently dead loops by checking the skillProposals flag and installed session skills. - Use Case: After a week of background sessions, run /skills-review to see three staged proposals, inspect one patch's diff, approve it into ~/.claude/skills/, and reject another with a recorded reason. ## Quick Start Ask the agent to list and review the pending skill proposals, then approve or reject each one after seeing its content.

Frequently Asked Questions about skills-review

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review pending skill proposals in Claude?

Run /skills-review with no arguments to list everything staged in ~/.claude/skills-pending/, including new skills, patches, and archive requests with their rationale. Then use show, approve, or reject with a proposal name to settle each one.

How do I approve a staged skill patch safely?

Use /skills-review approve <name>, which first shows a real diff generated by patch_apply.py and asks for confirmation before applying. The patch is applied atomically, logged to ~/.claude/skills-applied.log, and the proposal file is removed.

Why does skills-review say no skill has ever been approved?

The health scan only detects new skills carrying origin: agent-proposed frontmatter, so approved patches never appear in it. Check ~/.claude/skills-applied.log instead, which records every applied patch and shows whether the learning loop is actually landing changes.

What happens when a patch proposal is stale?

If an old_string anchor no longer matches the target file, patch_apply.py reports the patch as stale and the review stops rather than guessing a new anchor. The recommended action is to reject the proposal and have it re-proposed against the current content.

Can a skill be archived if other skills still reference it?

No. Before archiving, the review checks that absorbed_into content has landed in the umbrella skill and that the Dependency check section shows no remaining references. If either check fails, the archive is refused until the references are fixed.