ai-tool-designer

# AI Agent Tool Designer

## Overview

This skill provides comprehensive guidance for designing tools that AI agents can use effectively. Whether building custom agent tools or any AI-accessible interfaces, these principles maximize agent success in accomplishing real-world tasks.

Note: Use the more specific mcp-builder skill if you want to create an MCP server.

The quality of a tool system is measured not by how comprehensively it implements features, but by how well it enables AI agents to accomplish realistic, complex tasks using only the tools provided.

---

## Agent-Centric Design Principles

Before implementing any tool system, understand these foundational principles for designing tools that AI agents can use effectively:

### 1. Build for Workflows, Not Just API Endpoints

**Principle:** Design thoughtful, high-impact workflow tools rather than simply wrapping existing API endpoints.

**Why it matters:** Agents need to accomplish complete tasks, not just make individual API calls. Tools that consolidate related operations reduce the number of steps agents must take and improve success rates.

**How to apply:**
- Consolidate related operations (e.g., `schedule_event` that both checks availability and creates the event)
- Focus on tools that enable complete tasks, not just individual API calls
- Consider what workflows agents actually need to accomplish, not just what the underlying API offers
- Ask: "What is the user trying to accomplish?" rather than "What does the API provide?"

**Examples:**
- ❌ Bad: Separate tools `check_calendar_availability`, `create_calendar_event`, `send_event_notification`
- ✅ Good: Single tool `schedule_event` with parameters for checking conflicts and sending notifications

### 2. Optimize for Limited Context

**Principle:** Agents have constrained context windows - make every token count.

**Why it matters:** When agents run out of context, they fail to complete tasks. Verbose tool outputs force agents to make difficult decisio