A specialized utility to identify agent token waste, redundant LLM calls, and risky routing changes before they impact your cloud billing.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-cost-eval-kit
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-cost-eval-kit 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-cost-eval-kit?
The agent-cost-eval-kit is a read-only analysis tool designed to audit AI agents for token waste, repeated tool invocations, and risky routing configurations. In modern AI operations, misconfigured workflows can lead to rapid cost escalations. By using this kit from the ecosystem of Openclaw Skills, developers can preemptively signal when and where an agent is consuming excess resources without producing downstream value.\n\nThis utility evaluates the efficiency of your AI setups using structured, evidence-based checks rather than making hasty decisions from a single noisy run. By keeping your operational data secure and read-only, it provides an objective health check on your agents' execution paths, giving you actionable statuses like Rollback Candidate or Investigate One Path before costs spiral out of control.
agent-cost-eval-kit Use Cases
- Identifying AI agents that are consuming excessive tokens relative to the value of their outputs.\n- Detecting infinite loops, redundant retries, or duplicate scheduled jobs in agentic workflows.\n- Assessing the cost impact of changing models or agent routing paths.\n- Establishing safety baselines before implementing production updates to agent architectures.\n- Selecting safe candidates for configuration rollbacks based on token consumption patterns.
How agent-cost-eval-kit Works
- Trigger Check: The user initiates an audit using commands such as
check token waste for [My_Agent].\n2. Analyze Execution Patterns: The skill runs non-intrusively to check recent runs, agent loops, and token utilization signals without needing private API credentials.\n3. Assess Risks and Repetitions: It evaluates potential loops, duplicate calls, or empty signals (e.g., jobs returning empty replies but consuming full loops).\n4. Generate Status Labels: It assigns a standardized status label (such as 'Investigate One Path' or 'Watch Only') to prioritize issue triage.\n5. Deliver Actionable Recommendation: Provides clear next steps (such as read-only verification or path reduction) to optimize agent efficiency.
agent-cost-eval-kit Setup
To get started with agent-cost-eval-kit from the collection of Openclaw Skills, you can install it globally or locally.\n\n### Installation\n\nbash\n# Standard local installation\nopenclaw skills install agent-cost-eval-kit\n\n# Global installation\nopenclaw skills install agent-cost-eval-kit --global\n\n# Force global installation to overwrite previous versions\nopenclaw skills install agent-cost-eval-kit --global --force\n\n\n### Activation and Usage\n\nYou can invoke the evaluation kit using any of the following natural commands:\n\nbash\neval cost change for [My_Agent]\neval token change for [My_Agent]\ncheck token waste for [My_Agent]\n
agent-cost-eval-kit Data Schema & Taxonomy
The agent-cost-eval-kit produces structured outputs with metadata indicating the severity and location of LLM waste. Below is the taxonomy used to catalog evaluation findings:\n\n| Output Property | Description | Example Values |\n| :--- | :--- | :--- |\n| Status | Actionability level indicating the urgency of the intervention | Investigate One Path, Watch Only, Rollback Candidate |\n| Fix First | The specific agent, job, or path identifying the main source of waste | daily_empty_signal_reminder |\n| Reason | Logical explanation of why the signal flagged this workflow | Recent runs repeatedly returned empty downstream actions |\n| Evidence | Quantitative observations gathered during the audit | Recurring runs, static tokens consumed |\n| Risk Level | Classification of potential cost/performance impact | Low, Medium, High |\n| Next Action | Remediation instructions to execute manually | Do read-only verification before pausing the job |
agent-cost-eval-kit Advanced Features
- Non-Intrusive, Read-Only Security: This tool executes without editing configurations, disabling cron tasks, or needing access to private API keys or sensitive user data.\n- Evidence-Based Evaluation: Avoids false positives from isolated high-cost runs by tracking cumulative metrics and structural patterns.\n- Granular Status-Driven Workflows: Integrates cleanly into developer triage pipelines by categorizing agent health into six clear, action-oriented labels.\n- Proactive Routing Risk Analysis: Signals unexpected cost spikes caused by model updates or micro-routing structural changes before they reach your monthly bill.
SKILL.md
Loading
Related Openclaw Skills
Cognitive Self-Evolution Engine
A local, zero-cost self-optimization framework for OpenClaw AI Agents that analyzes execution logs to diagnose capability shortfalls and safely apply behavioral patches.
Market Insights & Competitor Monitoring Report Generator
An automated market intelligence and price monitoring agent that scrapes, scores, and delivers competitor reports with Alipay-enabled paywalls.
Mozilla Thunderbird Mail Storage Reader
An offline processing skill designed to inspect, search, and extract emails and attachments directly from local Mozilla Thunderbird storage profiles.
Broadband Grant Application Drafter
An AI-powered interview and drafting assistant that builds structured grant application narratives for federal and state broadband funding programs like BEAD, USDA ReConnect, and E-Rate.
Douyin Scraper
An automated Playwright-based scraper designed to harvest trending Douyin video metadata, engagement statistics, and copywriting scripts.
Tianshu Script Analyzer (天枢剧本库)
An AI-powered analytical engine for dissecting the viral success factors of domestic short dramas, offering comprehensive multi-dimensional reports and reusable creation formulas.
Featured*
