A structured episodic memory system that uses Q-value scoring to help AI agents remember successful patterns and avoid past mistakes.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install guava-memory
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install guava-memory using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is GuavaMemory?
GuavaMemory is a sophisticated episodic memory framework designed to provide AI agents with long-term learning capabilities within the ecosystem of Openclaw Skills. Unlike simple logging, this system utilizes a reinforcement learning approach by assigning Q-values to specific task outcomes. This allows agents to distinguish between efficient solutions, roundabout successes, and outright failures.
By documenting every task as a unique episode, the agent builds a searchable database of experience. The core value of this skill lies in its ability to transform raw execution history into actionable intelligence, ensuring that agents become more effective the more they are used. It bridges the gap between static instructions and dynamic, experience-based execution.
GuavaMemory Use Cases
- Reducing token usage and latency by retrieving successful procedures from previous tasks instead of re-solving problems.
- Building a robust library of anti-patterns to ensure Openclaw Skills do not repeat known logical errors or tool failures.
- Automatically promoting recurring high-success workflows into formalized skills for more reliable agent behavior.
- Monitoring agent performance trends through Q-value analytics and success/failure pattern tracking.
How GuavaMemory Works
- The agent initiates a memory search at the start of any new task to find the top 3 related episodes based on semantic similarity and Q-value.
- Upon task completion, the agent records a detailed episode file containing the intent, context, and specific success or failure patterns observed.
- A reward is calculated based on execution efficiency (e.g., 1.0 for one-shot success, 0.7 for success with trial and error).
- The system updates the Q-value of the task using a learning rate formula: Q_new = Q_old + 0.3 * (reward - Q_old).
- If a task intent achieves a high Q-value across multiple attempts, the system triggers a skill promotion to move the procedure into a permanent skill library.
GuavaMemory Setup
To deploy GuavaMemory for your Openclaw Skills, follow these configuration steps:
- Initialize Workspace Directories:
mkdir -p memory/episodes memory/skills memory/meta
Create the Index File: Initialize
memory/episodes/index.jsonto track system statistics and configuration thresholds.Configure Agent Rules: Update your AGENTS.md with specific episodic memory rules to enforce mandatory memory searches at task start and recording at task completion.
Install Search Utilities: Ensure
jqis installed on your system to enable the provided shell-based search scripts for manual episode auditing.
GuavaMemory Data Schema & Taxonomy
GuavaMemory uses a standardized schema to ensure all Openclaw Skills data remains portable and searchable:
| Component | Storage Type | Description |
|---|---|---|
| Episode MD | Markdown | Detailed logs including Intent, Success Patterns, and Failure Patterns. |
| Metadata Index | JSON | Central registry for episode IDs, Q-values, and promotion status. |
| Utility Metrics | Key-Value | Tracks rewards (0.0 to 1.0) and 'feel' (flow, grind, frustration, eureka). |
| Skills | Markdown | Optimized, graduated procedures that have passed the Q-value threshold. |
GuavaMemory Advanced Features
- Integration with Voyage AI for high-accuracy semantic search across all stored episodes.
- Automated Skill Promotion: Moves successful patterns to a dedicated skills directory once they meet the 0.85 Q-value threshold.
- Anti-Pattern Tracking: Maintains a dedicated log of critical failures to serve as a negative constraint for future planning.
- Dynamic Q-Value Decay and Updates: Implements a learning rate to prioritize recent experience over outdated task patterns.
SKILL.md
Loading
Related Openclaw Skills
GuavaGuard
A real-time runtime security monitor that identifies and warns against malicious patterns during agent execution.
whoopskill
A comprehensive CLI tool for fetching and analyzing WHOOP health metrics including sleep, recovery, HRV, and strain.
Canvas OS
An application platform that turns the OpenClaw Canvas into a dynamic environment for running and controlling rich HTML/CSS/JS applications.
Antigravity Image Generation
Native high-quality image generation using the Google Antigravity API and Gemini 3 Pro without browser automation.
GuavaSuite
A premium 2-layer security suite for AI agents featuring $GUAVA token-gated strict protection and on-chain identity verification.
Memory Mastery
A sophisticated three-layer memory architecture that provides AI agents with persistent context and semantic search capabilities across sessions.
Featured*
