cursor-subagent-creator

To create Cursor subagents, you generate `.cursor/agents/<kebab-case>.md` files with specific YAML frontmatter and prompt structure. This isolates context for specialized agents like verifiers or debuggers, enabling them to operate without interfering with the parent Cursor Agent.

You need isolated context for AI agent delegation when building specialized assistants like auditors or workflow orchestrators that require focused execution. It prevents interference between agents during complex multi-step workflows by giving each specialist a dedicated prompt and environment.

Yes, you can run background execution for long-running subagents in Cursor by setting the `is_background` frontmatter field to true. This enables asynchronous delegation for long-running streams, contrasting with synchronous foreground execution for sequential steps.

Cursor supports automatic routing for delegated subagents by using agent-friendly frontmatter fields like `description`, `model`, and `readonly`. The parent Cursor Agent reads these routing signals to automatically decide when to delegate tasks to the appropriate specialized agent.

The best way to structure a Cursor subagent markdown file is using kebab-case naming in the `.cursor/agents/` directory, incorporating YAML frontmatter with `description`, `model`, `readonly`, and `is_background` fields, followed by a clear prompt structure defining the agent's specialized behavior.