Refines and corrects stored AI preferences by decreasing confidence scores and facilitating memory updates.
The ccmem-correct skill is a specialized utility for managing the ccmem ecosystem within Claude Code. It allows users to provide negative feedback on specific AI memories, reducing their confidence levels to prevent the AI from relying on outdated or incorrect information. By automatically transitioning memories to 'under_review' or 'archived' status based on confidence thresholds, it ensures the AI context remains high-quality. Additionally, it streamlines the workflow by prompting for replacement information immediately after a correction is made.
주요 기능
01Incremental confidence reduction (-0.2) for targeted feedback
020 GitHub stars
03Seamless integration with replacement memory creation workflows
04Automated lifecycle management including review and archival states
05Clear visual feedback of confidence score transitions
06Negative reinforcement tracking for long-term behavior tuning
사용 사례
01Correcting misunderstood coding patterns or architectural guidelines
02Pruning low-value or incorrect context from the AI's long-term memory
03Deprecating outdated project preferences like switching from npm to pnpm
