tool-use

Name: tool-use
Availability: InStock
Author: xoai

Select and execute external tools across local scripts, MCP proxy, and subagent dispatch layers.

25|7|Updated Mar 16, 2026

Search Tags:#context #llm #agent #mcp #efficiency #tools

Ready to run

npx skills add https://github.com/xoai/sage --skill tool-use

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: tool-use
Source: https://github.com/xoai/sage/tree/main/core/capabilities/context/tool-use
Command: npx skills add https://github.com/xoai/sage --skill tool-use

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill ensures that AI agents select the most cost-effective and appropriate tools for external information retrieval or task execution, preventing the main context window from being overwhelmed by raw tool outputs.

Core Features & Use Cases

Layered Tool Selection: Prioritizes local scripts (zero cost), then MCP proxy (minimal cost), and finally subagent dispatch (isolated context) for complex tasks.
Context Management: Prevents pollution of the main AI context with verbose tool outputs, ensuring efficient token usage.
Use Case: When an agent needs to verify a code import, it first tries a local bash script. If it needs specific, up-to-date documentation, it uses an MCP proxy. For complex research requiring multiple steps, it dispatches a subagent.

Quick Start

Use the tool-use skill to find the current documentation for Next.js server components.

Frequently Asked Questions about tool-use

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I stop LLM agent context windows from filling up with verbose tool outputs?▼

To stop LLM agent context windows from filling up with verbose tool outputs, use a layered tool selection approach that isolates tool-generated data via subagent dispatch, preventing main context pollution and ensuring efficient token usage.

What is the best way to manage external tool execution costs for AI agents?▼

The best way to manage external tool execution costs for AI agents is prioritizing a three-layer selection system: starting with zero-cost local scripts, moving to minimal-cost MCP proxy, and finally dispatching isolated subagents for complex research.

How do I select the right tool for multi-step research without exceeding token limits?▼

To select the right tool for multi-step research without exceeding token limits, dispatch a subagent to handle the complex task in an isolated context, retrieving only the final verified results back to the main agent.

When should I use an MCP proxy instead of local scripts for tool selection?▼

You should use an MCP proxy instead of local scripts when an agent needs specific, up-to-date external documentation that requires minimal cost but cannot be verified locally, reserving subagent dispatch for more complex multi-step tasks.

Does dispatching a subagent for tool use isolate the raw data from the main context?▼

Yes, dispatching a subagent for tool use isolates the raw tool-generated data from the main context, satisfying the requirement for efficient resource utilization by preventing verbose outputs from overwhelming the primary LLM context window.