self-improving-agent

You record errors and corrections by triggering the logging behavior in Minis, which automatically writes entries into LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md to centralize reusable insights for cross-task improvement.

The best way to record command failures and outdated knowledge for traceability is to centralize entries into structured logs like ERRORS.md and LEARNINGS.md, enabling cross-task improvement and closing learning loops.

Yes, logging learning insights across tasks requires a valid SKILL.md frontmatter containing a name and description, while optionally using scripts, references, or assets folders to implement logging and memory integration.

Yes, you can expand the storage of learning logs to the project level, extending beyond the default skill data area categorization of LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md to structure traceability across broader contexts.

Centralizing external API issues drives structured cross-task improvement by capturing errors and corrections into shared logs like LEARNINGS.md, closing learning loops and providing traceability for future Minis operations.