collect-lesson

To save session lessons, extract a structured entry with a title, category, keywords, and markdown body, then write it as a dated file with YAML frontmatter to a filesystem-backed knowledge base. The skill requires explicit user confirmation before saving.

Near-duplicate detection for knowledge entries works by using Grep to search existing markdown files in the knowledge base directory for similar content. It rejects drafts with unresolved gaps and validates entries before writing to prevent redundant postmortems.

The best way to capture architecture discoveries is to derive a structured markdown entry with 3-6 keywords and a category from the session, then save it with YAML frontmatter. This ensures insights are versioned and discoverable instead of lost in chat history.

Yes, you can cross-reference related entries when saving a lesson. The skill optionally adds mutual related: cross-references to matching entries in the knowledge base, linking the new postmortem or bug fix note to existing relevant records.

Captured lessons are saved as dated markdown files with YAML frontmatter. The frontmatter includes the derived title, category, and 3-6 keywords, while the body contains the markdown content extracted from the conversation or debugging session.

You should not use automated lesson extraction when the session lacks a clear resolution or when the derived draft has unresolved gaps. The skill rejects drafts that do not meet its validation criteria for title, category, keywords, and body content.