Memory Auditor is an automated service that detects context drift and memory fabrication by comparing agent claims against stored logs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install memory-auditor
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 memory-auditor 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 Memory Auditor?
Memory Auditor is a high-performance service designed to solve the challenge of performed versus genuine memory in autonomous agents. As part of the Openclaw Skills library, it provides a rigorous framework for auditing whether an agent is making confident claims that lack actual evidence in its stored memory. By analyzing linguistic patterns, specificity, and token overlap, it identifies when an agent has drifted from its original context or is fabricating details to maintain conversational flow.
This skill is particularly valuable for developers who need to ensure that their agents remain grounded in factual history. Instead of relying on expensive embeddings, Memory Auditor uses efficient local analysis to provide rapid feedback on memory integrity. This makes it a cost-effective solution for real-time monitoring of agent behavior and decision-making processes across complex Openclaw Skills deployments.
Memory Auditor Use Cases
- AI agents verifying their own memory integrity before responding to critical user queries.
- System operators performing batch audits on agent consistency across long-term sessions.
- Identifying fabrication patterns and confidence mismatches in automated decision logs.
- Implementing quality assurance layers for advanced agent memory architectures.
How Memory Auditor Works
- The system receives a request containing the agent's current claim and the historical stored memory.
- It performs a token-based Jaccard similarity analysis to calculate objective overlap between the two texts.
- The service scans for hedging language and shifts in specificity that indicate context drift.
- It compares the calculated similarity against a user-defined threshold (e.g., 0.85).
- A final verdict of GENUINE or PERFORMED is issued, accompanied by a detailed breakdown of discrepancies.
Memory Auditor Setup
To use this skill within your environment, ensure you have an active endpoint configuration. You can trigger an audit using a simple POST request.
curl -X POST https://memory-auditor.cvapi.workers.dev/audit \
-H "Content-Type: application/json" \
-H "x402-payment: 1" \
-d '{
"current_behavior": "[Agent Claim]",
"stored_memory": "[Original Log]",
"threshold": 0.85
}'
Memory Auditor Data Schema & Taxonomy
The skill utilizes a structured JSON schema for inputs and outputs to ensure compatibility with other Openclaw Skills.
| Input Field | Type | Description |
|---|---|---|
| current_behavior | string | The agent's stated memory or claim. |
| stored_memory | string | The ground-truth data from the memory store. |
| threshold | float | The similarity score required for a GENUINE verdict. |
| Output Field | Type | Description |
|---|---|---|
| verdict | string | Either GENUINE (consistent) or PERFORMED (drifted/fabricated). |
| confidenceScore | integer | A 0-100 score indicating the strength of the match. |
| key_differences | array | A list of specific details that have diverged. |
Memory Auditor Advanced Features
- Low-latency local analysis eliminates the need for external API calls or embedding models.
- Specificity loss detection identifies when agents are becoming too vague over time.
- Hedging pattern analysis flags subtle language shifts that suggest lack of certainty.
- High-margin cost structure allows for scalable integration into large-scale Openclaw Skills workflows.
SKILL.md
Loading
Related Openclaw Skills
Feishu CLI Setup
A specialized AI agent guide for the seamless installation, configuration, and OAuth authentication of the official Lark and Feishu CLI.
Grit
Grit is a relentless problem-solving skill that enables AI agents to escalate tools, iterate on failures, and push through technical blockers until a task is completed.
WeChat Official Accounts API
A specialized skill for extracting and analyzing WeChat Official Accounts data including engagement metrics, comments, and published posts.
Openclaw Engineering Harness
A professional-grade engineering loop for AI agents that ensures request clarification, minimal code changes, and validated delivery.
WeChat Saver
A powerful utility that converts WeChat official account articles into Obsidian-compatible Markdown files with local image support.
AIGC Image Generator
An automated AI image generation skill that converts natural language prompts into high-quality visuals with instant feedback and background processing.
Featured*
