token-optimizer

What This Skill Does

The token-optimizer skill is a comprehensive toolkit designed to drastically reduce OpenClaw operational costs and token consumption. By implementing smart model routing, automated context selection, and efficient heartbeat management, this skill solves the 'context bloat' issue where agents load excessive files unnecessary for simple tasks. It uses intelligent classification to determine the minimum required context, reducing API overhead by 50-80%.

Installation

To integrate this skill into your environment, run the following command in your terminal: clawhub install openclaw/skills/skills/dexiaong/token-optimizer Once installed, ensure you initialize the heartbeat template by copying assets/HEARTBEAT.template.md to your ~/.openclaw/workspace/HEARTBEAT.md file to begin tracking and optimizing your agent's active lifecycle.

Use Cases

This skill is perfect for developers managing agents at scale who are hitting API rate limits or facing high billing costs. It is specifically designed for:

• Large Scale Deployment: Managing multiple agents where minor token savings compound across thousands of calls.
• Cost Control: Capping expenditure for hobbyist or enterprise budgets by switching to lower-cost models like Haiku for casual interactions.
• Performance Optimization: Improving agent response latency by minimizing the volume of data loaded into the prompt context during initialization.

Example Prompts

1. 'Run the context optimizer check to see how much I can save on my next session.'
2. 'Recommend the minimum files needed to answer: How do I refactor the database module?'
3. 'Check my current token budget status and switch to the most cost-effective model for this chat.'

Tips & Limitations

The biggest performance boost comes from running generate-agents to create an optimized AGENTS.md file. Always review the generated optimized files before deploying to production. Note that aggressive context reduction may occasionally lead to missing background information for highly niche, multi-step tasks; use the 'full' context mode when performing complex architectural analysis. Regular updates to your HEARTBEAT.md are required to maintain accurate usage metrics.