What problem does it solve? Agents with persistent project memory often accumulate stale, speculative, or repeated-wrong findings that erode trust in their reviews. This Skill defines how to write, organize, and calibrate that memory so it stays accurate and actionable across sessions. ## Core Features & Use Cases - Memory layout: A MEMORY.md index (max 200 lines, one line per topic) plus one Markdown file per topic, organized semantically rather than chronologically. - Entry format: Each entry records a claim, its reasoning (Why), an operational consequence (How to apply), a type, and a date. - Three entry types: pattern (defects to flag), false-positive-suppression (wrong findings to stop raising), and fixed-with-guard (resolved issues watched for regression). - Use Case: After a reviewer repeatedly flags a module-level cache as a concurrency bug in a single-worker runner, record a false-positive-suppression entry so the finding is never raised again. ## Quick Start Ask the agent to record a recurring review finding in project memory using the memory-discipline entry format with its Why and How-to-apply lines.
