ki-agents

You can audit Claude Code subagent definitions by running the audit mode against a rubric covering frontmatter formatting, name uniqueness, description length, and system prompt checks. It also requires human judgment for nuanced aspects like role and lane.

The best way to write a Claude Code subagent definition is to scaffold it using predefined guidelines for best practice and structure. This ensures proper scoping for roles, lane, and tool requirements.

Subagent definition review works by applying automated checks for mechanical criteria like frontmatter formatting and name uniqueness, while requiring human judgment to evaluate nuanced aspects such as description, role, and lane.

Yes, you can use a comprehensive rubric to check Claude Code subagent frontmatter formatting, description length, and system prompts. The automated mechanical checks ensure your definitions adhere to structural best practices.

The limitation of automated subagent definition audits is that they only cover mechanical criteria like frontmatter and naming. Nuanced aspects involving description quality, role definition, and lane assignment require human judgment to evaluate properly.

Yes, you need an existing subagent definition file to start a Claude Code audit. You can initiate the audit process by running the audit mode with init against a specific file path like './agents/code-reviewer.md'.