Authoring Subagents

What problem does it solve?

This Skill empowers you to design and implement autonomous AI subagents that handle complex, multi-step tasks with precision and human oversight, automating workflows and reducing manual effort.

Core Features & Use Cases

• Ask-Before-Write Pattern: Design agents that propose actions and reasoning for approval, ensuring transparency and user control before execution.
• Model Selection Guidance: Learn to choose the optimal AI model (Haiku, Sonnet, Opus) for each subagent's task, balancing intelligence, speed, and cost.
• Use Case: Implement a "Code Reviewer" subagent that proactively analyzes pull requests, identifies potential bugs or style violations, and proposes fixes, awaiting your explicit approval before making any changes.

Quick Start

Define a new subagent with a clear role, specific tools, and a step-by-step workflow. For example, to create a code formatter:

name: code-formatter description: Format code according to project standards tools: Read, Edit model: sonnet

You are a code formatter ensuring consistent style. When you are invoked:

• User asks to format code or files
• Code needs style standardization Your workflow:

1. Read the target file
2. Identify formatting issues
3. Propose changes with reasoning
4. Ask: "Should I proceed with these changes?"
5. After approval: Apply edits
6. Read back and report results