agent-setup

Invoke the agent-setup skill and choose guided, inference, or batch mode. It generates the YAML frontmatter with name, description, tools, model, and color, plus all required sections like Core Responsibilities, Hard Constraints, and Self-Verification Checklist.

Create an agent for autonomous multi-phase workflows, chain-of-thought reasoning, self-correction, parallel execution, or large-scale analysis of 50 or more files. Create a skill for user-invoked commands, interactive workflows, and simple linear tasks.

The frontmatter must include name, description, tools, model, and color. The description must contain at least three usage examples, each with context, user message, assistant response, and a commentary block explaining why the agent fits.

Yes, the skill supports update mode. It reads the existing agent file, identifies missing sections such as Operational Guidelines or Self-Verification Checklist, and adds them while preserving your existing content.

Use opus for complex autonomous reasoning and critical decisions, sonnet for balanced technical writing and structured analysis, and haiku for simple fast tasks. The skill infers a recommendation based on your agent's purpose and lets you confirm or change it.

Add utility scripts when the agent performs parallel execution across 50 or more files, runs multi-phase orchestration, or generates structured JSON or YAML reports. Scripts in Python, PowerShell, or Bash reduce token usage and accelerate execution.