reflexion:memorize

To consolidate agent reflections into CLAUDE.md, this Skill transforms raw critique outputs and execution feedback into atomic, verifiable guidance bullets. It organizes these actionable entries by section to help future agents reuse verified patterns and avoid repeated mistakes.

Agentic context engineering for knowledge management is a curation process that applies rules like relevance, non-redundancy, atomicity, and verifiability. It grows and refines repository context files to prevent context collapse while maintaining stability and traceability.

Yes, you can preview proposed CLAUDE.md updates before writing through the dry-run feature. This allows you to review consolidated reflection bullets and run scoped updates to avoid overwriting or diluting existing repository context.

To avoid duplicating context when adding execution feedback, the Skill applies de-duplication and quality gates during memory consolidation. It runs actionability tests and consolidation reviews to ensure only non-redundant, verified patterns are written.

Yes, this memory consolidation approach works with existing critique reports. It is designed for agentic workflows that need to convert conversation outputs, reflection reports, and verification outcomes into organized CLAUDE.md entries for reuse.