Memory Hygiene

Maintain high reasoning fidelity and prevent context window degradation by systematically auditing, triaging, compacting, and archiving persistent agent memory stores (MEMORY.md, .cursor/rules, AGENTS.md), ingesting in-flight compaction checkpoints, verifying references against live disk truth, and mitigating OWASP ASI06 memory poisoning.

When to Use

• Starting work in a repository with existing persistent agent memory or project scratchpads
• Memory files (MEMORY.md, .agents/memory.md) exceed line or token budget thresholds (e.g. >100 lines, >1,500 tokens)
• Ingesting structured in-flight checkpoints from agent-memory-compaction into persistent disk storage
• Detecting stale architectural assumptions, obsolete microservice endpoints, or dead file paths
• Eliminating duplicate, overlapping, or conflicting rules accumulated across multiple agent sessions
• Quarantining unverified facts or untrusted web/tool outputs to prevent cross-session memory poisoning (OWASP ASI06)
• Preparing a clean, compact memory snapshot before major multi-agent workflow handoffs

Core Rules