skill-propose

Capture reusable procedures from a session as new skills or patches to existing ones.

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

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When a session teaches a reusable procedure — a corrected approach, a discovered workflow, a dead end worth avoiding — that knowledge is normally lost when the session ends. This Skill turns those lessons into persistent skills, either by patching an existing skill or staging a new one, so future sessions benefit without manual note-taking. ## Core Features & Use Cases - In-flight capture: Triggers proactively when the user corrects the same output twice, proposing a fix at the moment the defect is obvious rather than at session close. - Patch-before-create discipline: Lists existing skills, prefers adding a labeled subsection to an umbrella skill over creating overlapping micro-skills, and verifies patch anchors mechanically with patch_apply.py before proposing. - Gated application with undo: Shows the verbatim change for approval, applies it atomically on a clear yes, and appends every change to ~/.claude/skills-applied.log so it can be reverted; otherwise stages proposals in ~/.claude/skills-pending/ for /skills-review. - Use Case: After a user corrects your writing style twice in one session, the Skill quotes the offending passage in the existing style skill, shows the exact replacement text, and patches it on approval — logging the diff for later undo. ## Quick Start Ask the agent to propose a skill capturing the procedure this session just taught, or invoke /skill-propose with a hint about what to capture.

Frequently Asked Questions about skill-propose

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

FAQPage Schema
How do I capture a reusable procedure from a Claude Code session?

Invoke /skill-propose with an optional hint about what to capture. The Skill reviews the session for qualifying signals — user corrections, recovered errors, or discovered workflows — then stages a new skill draft or a patch to an existing skill for your approval.

When should I patch an existing skill instead of creating a new one?

Patch when an existing skill is the umbrella for what you learned; the Skill adds a labeled subsection rather than creating an overlapping micro-skill. Create a new skill only when no existing class covers the procedure.

Can a proposed skill change be applied without review?

Only on the fast path: when you are present and corrected the same output twice, the Skill shows the verbatim change and applies it on a clear yes. Every application is logged to ~/.claude/skills-applied.log so it can be reverted.

What happens to proposals when nobody is available to approve them?

They are staged in ~/.claude/skills-pending/ as draft SKILL.md files or patch documents. The /skills-review command later promotes or rejects them; nothing becomes active until explicitly approved.

Why does patch verification fail on some proposals?

The patch_apply.py checker refuses proposals whose old_string anchors do not match the target skill exactly once, or whose change blocks are wrapped in three backticks instead of five. Fix the anchors or fencing and re-run the check before presenting the proposal.