memory-consolidate

What This Skill Does

The memory-consolidate skill acts as a persistent cognitive layer for OpenClaw agents, solving the common problem of session-based amnesia. It functions as an automated garbage collector and semantic summarizer for your agent's interactions. By reading session logs, it extracts critical facts, decision-making rationales, and project solutions. It utilizes a temperature-based lifecycle model to manage the 'freshness' of information, decaying stale data while promoting high-signal information into a dynamically generated MEMORY_SNAPSHOT.md file. This snapshot is automatically injected into every new session, ensuring your agent maintains context, history, and continuity across time. It optimizes the Signal-to-Noise Ratio (SNR) by clustering related concepts and deduplicating redundant logs.

Installation

To install, run: clawhub install openclaw/skills/skills/blusehuang1121/memory-consolidate. After installation, execute mkdir -p $OPENCLAW_WORKSPACE/memory/structured/{archive,candidates,semantic} to initialize the storage directories. Patch your agent config to inject the snapshot by running the provided gateway config.patch command. Finally, schedule the daily consolidation task via cron to ensure your agent stays updated at 3:00 AM UTC. Verify the health of your memory system anytime using the observation script: python3 $OPENCLAW_WORKSPACE/scripts/memory_consolidate_observe.py.

Use Cases

• Long-term project tracking: Retain technical decisions and architecture choices over weeks of development.
• Agent persona maintenance: Ensure the agent remembers user preferences and identity traits consistently.
• Knowledge base refinement: Filter out chatter and surface actionable facts from daily log noise.
• System Recovery: Rapidly restore context if a session is lost or corrupted.

Example Prompts

1. "The agent seems to have forgotten our project structure; please trigger a memory consolidate to fix the missing details."
2. "My agent's memory health is low. Can you run the memory optimization and check the SNR?"
3. "Why does the agent keep asking for setup details I provided last week? Please run a memory snapshot update."

Tips & Limitations

• Configuration: Always ensure IDENTITY.md and USER.md are populated; the skill relies on these for context-aware extraction.
• Tuning: If you find the agent is holding onto too much irrelevant data, increase the temperature.age_lambda value in your config.yaml to speed up the decay of older items.
• Dependency: Ensure the tui provider in your openclaw.json is configured correctly, as the semantic clustering relies on this LLM connection to perform high-level synthesis of your logs.