memory-hygiene

Memory Hygiene is a specialized utility designed for Clawdbot users to maintain a lean, efficient vector database (LanceDB). As AI agents accumulate information through auto-capture, vector memories often become cluttered with transient logs, repeated heartbeats, and low-value data that increase token costs during RAG (Retrieval-Augmented Generation) recalls. This skill provides a structured toolkit to audit current memory usage, perform surgical deletions, and establish automated cron-based maintenance routines to ensure only high-impact facts, preferences, and architecture decisions remain indexed.

• Perform deep audits of LanceDB memory contents using targeted recall queries.

• Execute complete wipes of the local memory directory to resolve persistent bloat.

• Re-seed essential data from master files like MEMORY.md to rebuild a clean state.

• Apply configuration patches to disable high-noise features like autoCapture.

• Automate monthly maintenance via cron jobs to parse, clean, and consolidate knowledge.

• Implement intelligent storage guidelines for distinguishing between transient state info and critical long-term facts.

• Users should monitor token usage; high consumption during simple queries often indicates a bloated vector memory.

• The tool is intended for developers and power users managing Clawdbot instances who want to optimize performance and reduce latency.

• Inputs include shell commands for directory management and JSON config patches for the gateway; outputs are refreshed memory states and improved agent recall precision.

• Always verify critical information in MEMORY.md before performing full wipes to prevent the loss of project architecture or credentials.

• Avoid storing heartbeat status, raw logs, or transient timestamps to keep the vector space reserved for high-importance knowledge.

• Use the provided importance weighting (0.7-1.0) when re-seeding to ensure optimal retrieval prioritization.