self-improvement

To log errors and corrections for continuous improvement, you capture events like user corrections and command failures, then write them to dedicated markdown files in a .learnings/ directory to preserve knowledge across sessions.

The best way to capture knowledge gaps and command failures is logging them directly into markdown files. This records non-obvious findings and external tool failures for later post-mortem analysis.

You promote broadly applicable learnings to project memory by elevating captured corrections and knowledge gaps into instruction files like CLAUDE.md, AGENTS.md, or Copilot instructions.

Yes, you can use self-improvement logging to update Copilot instructions and AGENTS.md. Broadly useful learnings captured during sessions are promoted directly into these project memory files.

You need a .learnings/ directory to start logging learnings and corrections. The implementation uses this dedicated directory alongside a standardized SKILL.md frontmatter to organize captured events.

You record external tool failures and user corrections during a session to prevent repeating mistakes. Capturing these events enables post-mortem analysis and promotes broadly useful knowledge to project memory.