task_spawn — Claude Code subagent: setup and task splitting
The task_spawn Claude Code subagent starts a separate agent for part of an active assignment. The new agent joins the same to-do list and reports to the same supervisor.
MIT ★ 395 updated 2026-11-16
.claude/agents/task_spawn.mdWhat task_spawn does
task_spawn starts a separate coding agent to handle part of the work already assigned to its parent. The child stays in the existing to-do and under the same supervisor; it does not become a team lead or open a new task.
Give it a short title and a self-contained prompt describing the work and what evidence to return. Unrelated work must go to the supervisor. The agent can replace a runtime’s standalone new-conversation action, such as Codex create_thread. The server derives the task, parent, launcher, and permissions from the verified assignment. When permission is unknown, access remains read-only. Do not retry a launch if its outcome is uncertain.
What it helps with
- split part of an active assignment into separate agent work
- give an agent a self-contained task with evidence to return
- run parallel work under the same supervisor and to-do
Example prompts for the agent
- Handle the error-reporting part of the current assignment and return the changes with supporting evidence.
- Inspect the current module for validation issues, stay within this assignment, and list the affected areas.
How to install task_spawn
Manually: put the files in place
.claude/agents/task_spawn.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/task_spawn.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/task_spawn.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about task_spawn
How do I install task_spawn in Claude Code?
Place task_spawn.md at .claude/agents/task_spawn.md.
What does task_spawn do?
It starts a separate agent for part of the current assignment. That agent joins the existing to-do and keeps the same supervisor.
What permissions does the child agent get?
The server derives permissions from the verified assignment. If permission is unknown, the child remains read-only.
Source text
task_spawn.md · MIT
Start a separate coding Agent for part of your current supervised task, on your current computer and directory. It joins the SAME to-do with the SAME supervisor; this does not make you a task leader or create a new task. Use this instead of a runtime's standalone new-conversation tool (for example Codex create_thread). Title and prompt must describe work within the existing assignment; ask your supervisor for an unrelated task. The server chooses the task, parent, launcher and permission from your verified assignment, never from arguments. Unknown permission stays read-only. The result names the child session, to-do and supervisor. Do not retry a start with an uncertain result.
```json
{
"type": "object",
"properties": {
"title": {
"type": "string",
"description": "Short title for this part of the existing task"
},
"prompt": {
"type": "string",
"description": "Self-contained assignment within the current task; include what evidence to return"
}
},
"required": ["title", "prompt"],
"additionalProperties": false
}
```
© nwverse · MIT · nwverse/persona