self-improvement

To log agent errors and corrections, you append structured entries to ERRORS.md and LEARNINGS.md within a .learnings directory. This centralizes agent failures, user corrections, and feature requests to enable continuous improvement through standardized tracking.

Tracking feature requests and tool outages is best handled by appending structured entries to FEATURE_REQUESTS.md and ERRORS.md in a .learnings directory. This organizes events into standardized formats with summaries, details, and metadata for future audits.

To set up a learning directory, create a .learnings folder and initialize LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md templates. You then append structured entries after events like failures, user corrections, or feature requests to capture insights.

Yes, you can promote high-value logged learnings into reusable project memory. An extraction workflow elevates captured insights into CLAUDE.md, AGENTS.md, and Copilot instructions to continuously improve agent performance across sessions.

Yes, the continuous improvement workflow supports optional automation hooks. These hooks integrate with the logging process to automatically capture failures, corrections, and feature requests into the .learnings directory structure.