Optimizes Claude's long-term memory by auditing and rating the relevance of injected context and observations.
The Retrospective Evaluation skill establishes a critical feedback loop for the Engram memory system, allowing Claude to refine the quality of information injected into its sessions. By assessing observations based on global utility and project-specific relevance, the skill helps suppress noisy or redundant data while boosting high-value architectural decisions and patterns. This process ensures that Claude's context remains lean, relevant, and increasingly helpful over time, preventing context window bloat and improving the accuracy of AI-driven suggestions.
主要功能
01Integration with Engram MCP tools for seamless memory state updates
02Suppression mechanics to permanently exclude low-value or repetitive tool noise
03Automated memory boosting for high-value architectural patterns
04Dual-axis evaluation of observations based on global and project-specific utility
050 GitHub stars
06Summary reporting of memory quality via formatted evaluation tables
使用场景
01Performing weekly maintenance to prune irrelevant or redundant project observations
02Reducing context window noise by suppressing re-discoverable facts and tool logs
03Promoting local project successes into global coding standards for use across all repositories
