agent-development_skill

Overview

This skill guides creating autonomous agents for Claude Code plugins using markdown files with YAML frontmatter. It explains required and optional fields, system prompt structure, and deployment locations so you can author reliable, triggerable agents. Follow the conventions to control model selection, tool access, and activation conditions.

How this skill works

You define an agent as a markdown file whose frontmatter lists metadata (name, description, model, color, tools) and whose body is the system prompt. The description field teaches Claude when to activate the agent with a short "Use this agent when..." section and 2–4 concrete triggering examples. The system prompt uses second-person instructions, a stepwise process, an explicit output format, and quality standards.

When to use it

• When designing a new autonomous specialist for a plugin or project.
• When you need precise activation conditions so Claude picks the correct agent.
• When constraining tool access or model behavior for safety and consistency.
• When standardizing agent prompts across a codebase or team.
• When testing agent activation and interaction flows before release.

Best practices

• Keep the description explicit and include 2–4 triggering examples to ensure correct activation.
• Use the principle of least privilege: list only the tools the agent needs.
• Write the system prompt in second person and include a clear, stepwise process.
• Specify an output format and quality standards to make parsing and evaluation reliable.

Example use cases

• A code-review agent that activates when users ask for PR review or mention security concerns.
• A deployment-checker agent that runs when users ask about deployment readiness or CI status.
• A documentation writer agent that triggers when the user requests API docs or examples.
• A troubleshooting agent that activates on error messages or requests for debugging steps.