Creates, edits, audits, and deletes Claude Code agents. Use when working with agent .md files in agents/ directories, building or improving agents, testing agent behavior, or understanding agent structure and best practices. Do NOT use for SKILL.md files (use author-skill).
context: fork), or team member (persistent via Task tool + team_name with SendMessage/TaskList coordination). An agent can be dual-purpose. Details: references/execution-models.md.Read, Grep, Glob. Code modification: Read, Edit, Bash, Grep. Omit tools field only when full access is genuinely needed.description field. Include: what it does + when to use it + differentiation from similar agents. Vague descriptions = never invoked.subagent_type. Test with general-purpose + inlined prompt (see audit-agent.md Step 3 for testing pattern).{skill-base-dir} from this skill's announced base directory and pass absolute paths to reference files in subagent prompts. Details: references/execution-models.md.Read references/path-resolution.md and resolve {agents-dir} and {plugin-docs}. Announce context before proceeding.
What would you like to do?
| Response | Next Action | Workflow |
|---|---|---|
| 1, "create", "new", "build" | Ask: "What task should this agent handle?" | workflows/create-agent.md |
| 2, "edit", "improve", "modify", "update" | Ask: "Which agent?" | workflows/edit-agent.md |
| 3, "audit", "review", "check", "test" | Ask: "Which agent?" | workflows/audit-agent.md |
| 4, "delete", "remove", "deprecate", "retire" | Ask: "Which agent?" | workflows/delete-agent.md |
Read the matched workflow file and follow it exactly.