All activity
M. Fahrul Alwanleft a comment
Built this after running an honest eval on my own agent. Score was 5/100 — not because the agent was bad, but because logging mistakes isn't the same as learning from them. This is the fix. Would love feedback from anyone who installs it.
metaskillTeach your AI agent how to actually learn, not just log
Most AI agents log mistakes. They don't learn from them.
metaskill fixes this — forces 3-level correction per error (surface → principle → habit), scans past patterns before tasks, and captures wins not just failures.
Built after OpenClaw scored 5/100 on its own learning eval. 37 corrections logged. Same errors repeating.
clawhub install metaskill — zero deps, works on any OpenClaw instance.
metaskillTeach your AI agent how to actually learn, not just log