subagent-creator

To create isolated AI subagents, you design a root SKILL-style prompt with YAML metadata including name and description, followed by actionable invocation steps and expected output formatting to ensure context separation.

Agent delegation assigns specialized subtasks to isolated subagents. You need isolated context when running complex multi-step workflows to prevent cross-task interference and avoid contaminating the main agent's context window.

Structure a debugger or verifier subagent prompt by defining identity, invocation steps, expected outputs, and quality expectations, using kebab-case naming and explicit descriptions for automatic parent agent delegation.

Building a security auditor subagent requires YAML metadata fields including name and description, plus optional model and readonly fields, to ensure the parent agent can automatically identify and delegate the security auditing task.

The best way to prevent cross-task interference is designing each specialized assistant with isolated context windows, ensuring each subagent runs independently without contaminating the main agent's context during multi-step workflows.

A parent agent fails to delegate when the subagent lacks clear delegation-ready metadata. You must write explicit name and description frontmatter stating exactly when the parent agent should automatically invoke the subagent.