self-update

You can automate rule evolution by analyzing session lessons and promoting recurring patterns into project rules. The system validates patterns against criteria, creates files at .claude/rules/project-*.md, and archives promoted lessons for traceability.

Promoting lessons into project rules works by scanning lessons.md for recurring agent behavior patterns, validating them against project-rule criteria, writing them to .claude/rules/project-*.md, and archiving the original lessons to ensure traceability.

You should run a retrospective to improve agent behavior rules during weekly review cycles or on demand via explicit requests. This surfaces recurring workflow patterns and extracts reusable knowledge into structured rules or new skills.

Yes, you can extract recurring workflows into new skills automatically. When a specific pattern recurs across multiple solutions, the system identifies it and extracts the reusable workflow into an independent skill.

The best way to manage project-specific knowledge growth is implementing a continuous improvement loop that scans session notes, promotes stable patterns into .claude/rules/project-*.md files, and archives lessons to maintain end-to-end governance.

Yes, you need existing lesson files like lessons.md or session notes. The improvement cycle scans these sources to surface recurring patterns in agent behavior before validating and promoting them into project rules.