reflexion:memorize

To save agent execution insights into CLAUDE.md, you index reflections, critiques, and feedback into structured, actionable bullets categorized by domain and pattern. This process converts reflection outputs into durable rules with clear success criteria for future reuse.

The ACE grow-and-refine approach for context engineering is a cycle that evolves CLAUDE.md by applying structured memory curation to reflection outputs. It generates non-redundant, actionable guidance bullets to improve future agent performance over time.

You structure execution feedback into actionable rules by extracting memorizable insights from critiques and applying the ACE grow-and-refine cycle. This produces evidence-backed, verifiable bullets categorized by domain and pattern, complete with concrete examples and clear success criteria.

Memorizing reflection insights into a playbook specifically avoids secrets or sensitive data during the indexing process. It extracts evidence-backed, non-redundant rules from execution feedback to ensure the CLAUDE.md context remains secure and verifiable.

You can use reflections and critiques to automatically update agent context playbooks by running the memorize command. It consolidates these inputs into CLAUDE.md, creating a living, evolvable playbook that provides actionable guidance for subsequent tasks.

You need to consolidate execution feedback into a structured context playbook when a reflection highlights a recurring pattern. Converting this feedback into targeted, evidence-backed rules generates durable context that directly improves future agent performance.