An automated diagnostic tool to audit AI agent performance, calculate token costs, and provide model optimization recommendations for Openclaw Skills users.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-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 agent-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 Agent Audit?
Agent Audit is a comprehensive diagnostic utility designed to help users understand the financial and operational efficiency of their AI workflows. By scanning configurations and analyzing execution history, this skill identifies exactly where resources are being utilized effectively and where costs can be trimmed without sacrificing performance within the ecosystem of Openclaw Skills. Whether you are running complex coding agents or simple status checks, this tool provides the transparency needed to manage AI model sprawl.
It specifically targets the balance between model capability and task complexity, ensuring that users are not overpaying for high-tier models when more efficient alternatives are available. As part of the broader suite of Openclaw Skills, Agent Audit provides actionable insights through detailed markdown reports, enabling developers to make data-driven decisions about their agent infrastructure. The skill focuses on maximizing ROI by classifying tasks and suggesting model-task fit optimizations across various providers like Anthropic, OpenAI, and Google.
Agent Audit Use Cases
- Auditing agent configurations to find cost-saving opportunities.
- Analyzing cron job history to identify expensive or inefficient tasks.
- Mapping specific tasks to the most cost-effective model tiers (e.g., Haiku vs. Opus).
- Calculating ROI and monthly token spend across different model providers.
- Determining task complexity to ensure model-task fit and performance.
How Agent Audit Works
- Discovery phase scans the local configuration files to map agents to specific tasks and detect providers from model names.
- History analysis pulls the last seven days of cron and session history to calculate average token usage and success rates.
- Task classification categorizes workflows into Simple, Medium, or Complex tiers based on reasoning requirements and output patterns.
- Recommendation engine identifies potential model downgrades for simple tasks while maintaining high-tier models for critical work.
- Report generation creates a detailed markdown summary including potential savings and specific configuration change strings for Openclaw Skills.
Agent Audit Setup
To run a full audit of your setup, use the following command:
python3 {baseDir}/scripts/audit.py
For specific output formats or dry runs, you can use these additional options:
# Generate a quick summary only
python3 {baseDir}/scripts/audit.py --format summary
# Preview what would be analyzed without generating a report
python3 {baseDir}/scripts/audit.py --dry-run
# Save the report to a specific file path
python3 {baseDir}/scripts/audit.py --output /path/to/report.md
Agent Audit Data Schema & Taxonomy
The skill classifies data into complexity tiers to determine model recommendations. The schema for classification is as follows:
| Tier | Recommended Models | Criteria |
|---|---|---|
| Simple | Haiku, GPT-4o-mini, Flash | Short output (<500 tokens), repetitive patterns, health checks |
| Medium | Sonnet, GPT-4o, Pro, Grok | Medium output, reasoning required, research tasks |
| Complex | Opus, GPT-4.5, Ultra, Grok-2 | Long output, multi-step reasoning, coding, security reviews |
All Openclaw Skills data is processed locally to generate markdown reports containing agent breakdowns, cron job frequency, and monthly spend estimates.
Agent Audit Advanced Features
- Multi-provider support including Anthropic, OpenAI, Google, and xAI model families.
- Safety-first logic that explicitly prevents downgrading recommendations for critical coding or security-related tasks.
- Confidence scoring and risk assessment for every suggested model change.
- Read-only operation to ensure configuration stability while providing optimization insights.
- Detailed heuristics for task classification based on historical performance and token usage within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Web Design Guidelines
A specialized auditing tool that reviews frontend code against Vercel's Web Interface Guidelines for accessibility and UX excellence.
Vercel React Best Practices
A comprehensive performance optimization framework for React and Next.js based on Vercel Engineering guidelines.
Vercel Deploy
A specialized tool for deploying web applications and static sites to Vercel directly from an AI agent environment without requiring initial authentication.
Pinak Frontend Guru
A high-performance React and UI/UX audit agent that transforms frontend code into production-ready professional applications.
Job Hunter
A comprehensive end-to-end AI assistant designed to automate job searching, fit evaluation, and application material generation.
ISS Tracker
A lightweight utility to fetch the real-time geographic coordinates of the International Space Station.
Featured*
