self-improvement

You can automatically capture errors and corrections by using shell scripts and hooks that trigger after unexpected command failures or user inputs, logging the events to a centralized .learnings directory for continuous improvement.

Automated learning capture works by applying hooks after prompts and tool executions across sessions, detecting recurring patterns and user corrections, then logging them into project memory files to improve future performance.

The best way to log recurring command failures is to use a centralized .learnings directory with provided templates, recording errors and knowledge gaps triggered automatically by hook mechanisms after execution failures.

Yes, you can promote important captured learnings into project memory files such as CLAUDE.md, AGENTS.md, and copilot-instructions to disseminate patterns and improve future task execution.

No external dependencies are required. The continuous improvement hooks rely solely on simple shell scripts and .learnings templates to log entries and optionally promote them to project memory.

You should extract logged learnings into a separate skill when high-value learnings recur frequently across tasks, using the skill extraction workflow to convert repeated corrections into a reusable automated pattern.