subagent-creator

Create a markdown file with frontmatter containing a kebab-case name, a description explaining when to delegate, and optional model and readonly fields. The body defines the subagent's identity, steps to follow when invoked, and expected output format.

Use a subagent when the task is complex with multiple steps and requires isolated context, such as verification, debugging, or parallel workstreams. Use a skill for quick one-off actions or reusable procedures that do not need context isolation.

Write a specific description stating exactly when the subagent should be used, and include trigger phrases like "use proactively when" or "always use for". Vague descriptions like "use for general tasks" prevent automatic delegation.

This skill targets agent-agnostic subagent definitions and explicitly excludes Cursor-specific subagents. For Cursor, use the dedicated cursor-subagent-creator skill instead.

Common mistakes include vague delegation descriptions, overly long prompts, and creating too many unfocused subagents. Best practice is one clear responsibility per subagent, concise actionable instructions, and starting with two or three focused agents.