A specialized audit tool that rewrites your OpenClaw configuration to drastically reduce Anthropic API expenses through intelligent model routing and prompt caching.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install anthropic-cost-optimizer
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 anthropic-cost-optimizer 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 Anthropic Cost Optimizer?
The Anthropic Cost Optimizer is a high-impact utility within the Openclaw Skills ecosystem designed to combat rising API costs. Following Anthropic's 2026 billing changes, many developers are facing full API rates for the first time. This skill serves as an automated financial auditor for your AI agents, scanning your configuration files for expensive inefficiencies.
By analyzing five critical cost levers—including prompt caching, model routing, and thinking levels—the skill identifies where you are overspending. It doesn't just point out problems; it generates a fully optimized version of your config, providing a transparent diff and estimated monthly savings before applying any changes. It ensures you use the right model for the right task, such as routing simple triage to Claude Haiku while reserving Claude Opus for complex reasoning.
Anthropic Cost Optimizer Use Cases
- Reducing high monthly API bills from Anthropic.
- Transitioning from Claude subscriptions to direct API billing.
- Optimizing OpenClaw configurations for better token efficiency.
- Automating the implementation of prompt caching (cacheRetention) across multiple agents.
- Benchmarking current AI agent spend against optimized model routing strategies.
How Anthropic Cost Optimizer Works
- The skill loads current model rates and savings formulas from a pricing reference file.
- It locates the active openclaw.yaml or openclaw.json file across standard directory paths.
- It performs a comprehensive audit across five cost levers: Prompt Caching, Model Routing, Thinking Level, Context Window size, and Fast Mode.
- Each issue is flagged with a severity rating (High, Medium, or Low) and an estimated impact.
- An optimized configuration is generated, preserving user settings while annotating all cost-saving changes.
- The user reviews a detailed diff and estimated savings before confirming the file write operation.
Anthropic Cost Optimizer Setup
To use this optimizer within your environment, ensure you have the pricing reference available and call the skill via your OpenClaw agent:
# Trigger the optimizer by asking your agent:
"My API bills are too high, optimize my config."
# After the skill writes the new config, restart your gateway:
openclaw gateway restart
Anthropic Cost Optimizer Data Schema & Taxonomy
The skill interacts with the following data structures and files:
| File/Component | Description |
|---|---|
references/pricing.md |
Source of truth for model rates and savings formulas. |
openclaw.yaml/json |
The primary target for auditing and rewriting. |
cacheRetention |
Metadata tag added to model params (e.g., "short" or "long"). |
thinking |
Parameter scoped to specific agents to manage compute costs. |
context1m |
Boolean flag managed to avoid unnecessary surcharge pricing. |
Anthropic Cost Optimizer Advanced Features
- Intelligent Model Tiering: Automatically reassigns tasks to Haiku, Sonnet, or Opus based on agent role (Triage vs. Reasoning).
- Prompt Caching Logic: Implements
cacheRetentionstrategies that can reduce input costs by 60-90%. - Surcharge Prevention: Detects and removes the 1M context beta header from agents that do not require high-token windows.
- Subscription Detection: Recognizes if you are using
sk-ant-oat-*tokens and warns that certain caching features require a standard API key. - Annotated Diffs: Provides a line-by-line explanation of every optimization made for full developer transparency.
SKILL.md
Loading
Related Openclaw Skills
Lobster Subagent-Driven Development
A high-integrity development framework that dispatches fresh subagents per task with mandatory two-stage reviews to ensure spec compliance and code quality.
YouTube Downloader & Skimmer
A specialized tool to download YouTube videos and automatically slice them into segments based on chapters or custom timestamps.
Lobster Debugging
A rigorous 4-phase debugging framework for AI agents that mandates systematic root-cause analysis over temporary symptom patching.
Search First
A systematic research-before-coding workflow that prevents reinventing the wheel by prioritizing existing tools, libraries, and MCP servers.
AI Stock Investment Advisor
A comprehensive AI-driven advisory system that automates global stock data collection, multimodal technical analysis, and professional investment report generation.
Whiteboard Animation Generator
A specialized tool that converts static images into dynamic whiteboard hand-drawing and coloring animations.
Featured*
