metaskill – AI agent learning through structured error correction

Summary: metaskill enhances AI agents by enforcing three-level error correction and analyzing past patterns before tasks. It captures both failures and successes to improve learning, addressing the gap between logging mistakes and actual learning.

What it does

metaskill forces corrections at surface, principle, and habit levels for each error, scans previous patterns before new tasks, and records wins alongside failures. It integrates with OpenClaw instances without dependencies.

Who it's for

It is designed for users running OpenClaw AI agents seeking improved learning from errors beyond simple logging.

Why it matters

It solves the problem of AI agents repeating errors by enabling deeper, structured learning rather than just mistake logging.