self-improvement

You log learnings and errors by creating a dedicated `.learnings/` store using provided templates, which enforces progressive-loading to capture corrections, failures, and feature requests for continuous improvement.

Progressive-loading is a workflow that maintains entries in a `.learnings/` store, allowing agents to incrementally capture and recall mistakes without overwhelming context windows, fueling structured continuous improvement.

You promote broadly applicable learnings to CLAUDE.md or AGENTS.md using the promotion workflow, which scales captured knowledge across the codebase by elevating critical entries from the `.learnings/` store.

Yes, the logging approach works without external dependencies, relying solely on markdown files, a `.learnings/` directory, and built-in templates in `assets/` and `references/` to structure entries.

You need to capture corrections and feature requests during development, debugging, and user feedback workflows when agents repeat issues or lack a structured mechanism to recall past mistakes and improvements.

The best way to structure development learnings is using centralized logging with consistent templates in a `.learnings/` store, enabling progressive-loading and promotion of critical knowledge to project memory files like CLAUDE.md.