token-pilot

What This Skill Does

Token Pilot is a comprehensive optimization engine designed for OpenClaw AI agents to dramatically reduce token consumption while maintaining high performance. By implementing a suite of six immutable rules (R1-R6), it enforces smart file reading, surgical tool execution, and response brevity. Beyond basic constraints, it offers deep synergy with existing plugins like memory agents, search engines, and team-builder modules, ensuring that context is recycled rather than re-discovered.

Installation

To integrate Token Pilot into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/beyound87/token-pilot Ensure you have Node.js installed, as this is a cross-platform, pure Node.js utility.

Use Cases

• Large Codebase Navigation: Ideal for developers who need to explore deep repositories without flooding the context window with extraneous code.
• Cost-Sensitive Operations: Essential for users operating under strict token limits or high-latency LLM usage where unnecessary verbose output increases cost and wait times.
• Agent Orchestration: Perfect for multi-agent workflows where agents need to share context efficiently via local disk files rather than redundant prompt embeddings.

Example Prompts

1. "Token Pilot, run an audit on my current session configuration and tell me if I'm wasting tokens on file reads."
2. "I need to refactor the authentication module. Use Token Pilot rules to explore the dependencies and report back briefly."
3. "Execute a memory check to see if we've already solved the CSS layout issue before you start searching the project files."

Tips & Limitations

• Context Management: Always prioritize the use of .openclaw/context.md files for large projects to minimize the prompt size. Keep your SOUL.md files under 600 tokens for optimal agent performance.
• Read Limits: Respect the R1 limit of 30 lines for initial reads. If you are a developer, use the surgical offset/limit features to narrow down debugging efforts.
• Exceptions: Remember that while brevity is the goal, you should read project standards files fully when setting up context to prevent downstream errors that cost more tokens than the initial read.
• Automation: The rules are triggered automatically on session start. There is no need for configuration scripts unless you are performing a manual system audit.