A diagnostic observability skill that tracks and analyzes AI agent tool calls to distinguish between necessary actions and reflexive API invocations.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-audit
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 self-audit 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 Self-Audit?
Self-Audit is a specialized tool designed to bring deep introspection to AI agent workflows. While most observability stacks focus on measuring external call metrics like latency or success rates, this Openclaw Skills entry focuses on the intent and necessity of those calls. It helps developers identify whether an agent is making a tool call because it truly lacks the information or if it is simply acting out of a reflexive pattern.
By implementing this skill, you can significantly improve the efficiency of your AI agents. It provides a structured framework to evaluate the tool-call-to-value ratio, helping to reduce operational costs and minimize unnecessary network overhead. It is an essential component for developers who want to move beyond basic execution logs and into the realm of intelligent agent behavior analysis.
Self-Audit Use Cases
- Auditing AI agent decision-making to identify redundant API calls.
- Identifying opportunities for caching to reduce latency in repetitive workflows.
- Distinguishing between an agent's capability gaps and reflexive tool usage.
- Benchmarking the efficiency of different agent versions or models within Openclaw Skills.
- Reducing API costs by internalizing knowledge that the agent frequently requests externally.
How Self-Audit Works
- The agent records every tool invocation using the log command, including the specific tool, input parameters, and the reasoning behind the call.
- All data is stored in a dedicated audit directory to maintain a historical record of agent behavior.
- The analysis engine processes these logs to detect patterns, such as repeated calls for static data.
- An efficiency score is generated based on how many calls were deemed necessary versus those that could have been avoided.
- The skill provides actionable recommendations, such as suggesting specific data to cache or identifying where internal knowledge should be expanded.
Self-Audit Setup
To get started with Self-Audit in your environment, follow these installation steps to integrate it with your existing Openclaw Skills:
# Ensure the self-audit script is executable
chmod +x self-audit
# Log a test tool call to initialize the audit directory
./self-audit log --tool "weather_api" --input "NYC" --reason "user_query"
# Generate your first efficiency report
./self-audit analyze
Self-Audit Data Schema & Taxonomy
The Self-Audit skill organizes its data to provide clear insights into agent behavior. The following table describes the primary data components:
| Component | Type | Description |
|---|---|---|
audit/ |
Directory | Stores local logs of all tool invocations and metadata. |
self-audit |
CLI Tool | The primary interface for recording, analyzing, and scoring calls. |
Efficiency Score |
Metric | A 0-10 rating based on the necessity of calls in a session. |
Pattern Metadata |
Analysis | Identifies behavioral loops such as reflexive or thoughtful usage. |
Self-Audit Advanced Features
- Necessity scoring to flag calls that could be internalized or handled via local knowledge.
- Automated caching suggestions based on the frequency and variance of tool responses.
- Pattern detection algorithms that identify behavioral loops and reflexive usage triggers.
- Session comparison capabilities to track the optimization of Openclaw Skills over multiple iterations.
- Detailed suggestions for architectural improvements to reduce agent latency.
SKILL.md
Loading
Related Openclaw Skills
Group Director
A specialized video execution layer that converts chat-derived prompts into high-quality short videos for enterprise communication platforms.
AI Intelligent Knowledge QA
A sophisticated AI-driven Q&A system combining semantic search and GPT integration to automate FAQ management and knowledge retrieval.
EdrawMind Mind Map Generation
A powerful integration that transforms structured Markdown or natural language into professional, editable EdrawMind maps with custom layouts and styles.
GTM Meeting Prep
A professional-grade intelligence tool that generates tailored pre-meeting briefs, account research, and attendee insights for sales and growth teams.
Defuddle Web Content Extractor
A high-performance web content extraction engine that transforms cluttered HTML into clean, structured Markdown for AI agents and developers.
AkShare Fund Analysis
A quantitative data analysis tool for public funds, enabling deep insights into holdings, valuations, and performance tracking via AkShare.
Featured*
