self-improvement

Description

TRIGGER when the user gives substantive correction or feedback about agent behavior — corrects/rejects/clarifies your action or conclusion, states a principle ("don't do that", "prefer X", "always Z"), evaluates agent quality, proposes changes to instructions/agents/skills/memory/repo/workflow, or reminds you that this skill should have run (a reminder counts as feedback — invoke in the current turn). Diagnose what went wrong and write concrete edits to ~/claude-agent-instructions/. SKIP for neutral confirmation ("ok", "thanks", "yes do it") and for pure questions that do not evaluate your actions. A reusable SKILL.md agent skill by sthe0.

Community Metrics

Categories

• Skills

Tags

• skill
• claude-skill
• github
• en

Capabilities

• SKILL.md package

Information