memory-hygiene

What This Skill Does

Memory Hygiene is a specialized utility for the OpenClaw agent designed to maintain the integrity and efficiency of the LanceDB vector memory system. Over time, automated agents often accumulate 'digital junk'—transient states, heartbeat signals, and redundant logs—that bloat the vector database and degrade the relevance of auto-recall operations. This skill provides the necessary audit tools, cleanup procedures, and configuration strategies to ensure that the AI agent's long-term memory remains lean, relevant, and performant. By implementing systematic memory maintenance, users can reduce token consumption and improve the agent's decision-making capabilities.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/chenghaifeng08-creator/memory-hygiene-automaton Ensure you have appropriate permissions to modify the .clawdbot/memory/ directory before executing the wipe commands.

Use Cases

• Agent Performance Optimization: When you notice the AI is providing outdated information or recalling irrelevant details from previous sessions.
• Token Cost Management: Reducing the 'context window noise' that forces the agent to process unnecessary vectors during memory lookups.
• Maintenance Automation: Setting up recurring cron jobs to clear volatile data on a monthly basis, ensuring a clean slate for project updates.
• Data Privacy: Removing sensitive information or accidental logs that have been indexed into the vector store.

Example Prompts

1. "Perform an audit of my memory using memory_recall and show me the last 20 entries to see if we have accumulated any junk."
2. "Disable autoCapture in the configuration so the agent stops saving heartbeat status and transient messages to LanceDB."
3. "Set up a monthly maintenance cron job to wipe the memory folder and re-index the essential facts from my MEMORY.md file."

Tips & Limitations

• Precision Over Volume: Always aim for one concept per memory entry. Avoid bulk-loading raw logs as these are better suited for file-based storage than vector memory.
• Caution with Wipe: The rm -rf command for LanceDB is destructive. Always ensure your essential 'golden facts' are stored in a human-readable file (like MEMORY.md) before triggering a full wipe.
• Importance Scoring: When storing new facts, use high importance (0.8+) for core workflows and lower importance for transient preferences to allow for easier future pruning.