What problem does it solve?
This Skill streamlines authoring custom Claude Code subagents by providing a complete, correct playbook for producing a well-structured agent definition that reliably triggers, has the right tool permissions, and returns output in a predictable shape.
Core Features & Use Cases
- Agent definitions that actually work: Guides you to set the minimal frontmatter (
name, description) plus optional capabilities like tools, model/effort, memory, skills preloading, and isolation.
- Tool restriction and safety alignment: Helps you choose allowlists/denylists and understand how permission mode interacts with tool access so the agent can’t do the wrong thing.
- Correct agent body structure: Teaches the five-section system-prompt format (Identity, Execution, Output Format, Failure Conditions, Constraints) so downstream orchestration can parse results.
- Scoping and storage routing: Covers where the agent should live (project, user, or plugin distribution) and what that implies for sharing and security boundaries.
- Debugging under-triggering and shape errors: Provides a decision flow and failure-mode guidance for agents that fail to fire or return the wrong output shape.
Quick Start
Use the agent-creator skill to design a new subagent when you need an isolated worker to handle a recurring task, such as creating or editing a custom subagent under .claude/agents/ with the correct tool restrictions, model/effort, and a locked output format.
