self-improvement

To log AI agent corrections, you capture errors and knowledge gaps into markdown files within a centralized .learnings store. This enables continuous improvement by preserving context for later analysis and future sessions.

Logging learnings to markdown helps with promoting insights by enforcing YAML frontmatter consistency and workflow requirements. Broadly applicable learnings are reviewed and promoted to workspace guidance files like CLAUDE.md or AGENTS.md.

Yes, you can automatically prompt for learning logs after tool failures. The system provides hooks and scripts that trigger logging prompts after tool failures or corrections, reducing cognitive load during development.

The best way to preserve AI context across sessions is logging corrections and knowledge gaps to a centralized learning store. This stores logs under .learnings for later processing and potential extraction of reusable skills.

Yes, the learning log promotion process enforces YAML frontmatter consistency. It ensures workflow requirements are met before promoting broadly applicable learnings to workspace guidance artifacts like CLAUDE.md and AGENTS.md.

You should avoid a centralized learning store for AI self-improvement when dealing with highly sensitive data, as logs are stored in markdown files. It is also less suitable for ephemeral sessions where long-term context preservation is unnecessary.