Guava Memory

What This Skill Does

Guava Memory is an advanced episodic memory management system designed for the OpenClaw AI agent framework. It enables the agent to store, recall, and optimize its workflows based on historical performance. By utilizing Q-value scoring—a reinforcement learning concept—the agent learns to weigh past experiences, promoting highly successful task-sequences into reusable "skills" while cataloging anti-patterns to avoid repeating costly mistakes. The system ensures that your agent evolves from a generic helper into a specialized expert tailored to your specific project needs.

Installation

Installation is streamlined via the ClawHub ecosystem. Execute the following command in your terminal within the OpenClaw workspace:

clawhub install openclaw/skills/skills/koatora20/guava-memory

After installation, ensure you initialize the memory directory structure: mkdir -p memory/episodes memory/skills memory/meta. Finally, verify that your agent's AGENTS.md is updated with the provided episodic memory rules to ensure the agent maintains consistency across sessions.

Use Cases

• Repeated Task Automation: Perfect for recurring coding tasks, such as boilerplate setup or standardized debugging flows, where the agent learns the most efficient sequence of tools.
• Error Prevention: If the agent consistently encounters a specific error when interacting with a legacy API, Guava Memory records this as an anti-pattern, ensuring the agent uses an alternative approach in the future.
• Agent Optimization: Great for long-term project development where agent performance needs to improve as it gains deeper context regarding your codebase and coding standards.

Example Prompts

1. "Check your memory for past successful deployments to the production server and follow the same steps for this new update."
2. "Search for the failure pattern related to the recent authentication timeout and suggest a more resilient retry strategy."
3. "Refine your current task procedure by reviewing episodes where you felt 'eureka' to boost your overall Q-value for this project."

Tips & Limitations

• Data Integrity: Always keep the index.json file in sync with your episodic files to prevent broken pointers in the memory system.
• Promotion Threshold: The promotion threshold of 0.85 is a balancing act; if your agent is too rigid, consider lowering the threshold. If it promotes too much noise, increase it.
• Independence: This skill relies on manual file-system interaction and local processing. It does not require external database connections, making it a privacy-friendly, local-first solution.