task_decided — Claude Code subagent: what it does, how to install
The task_decided Claude Code subagent records an owner's answer to a pending task decision. It captures the choice when the owner gives it in the conversation.
MIT ★ 395 updated 2026-11-16
.claude/agents/task_decided.mdWhat task_decided does
task_decided saves the owner's answer to a task decision that is waiting in its to-do list. It accepts only the owner's words from the main conversation or the task's supervisor. An agent's own judgement does not count as the owner's decision.
The record needs the task ID, the exact label of the chosen option, and the owner's words. If the answer matches no option, the subagent asks for clarification or for the decision to be revised with task_review. With told_agent: false, the hub sends the selected option to the task agent and records it; if that agent cannot be reached, the record is refused. With told_agent: true, the answer has already been sent through agent_send, so the subagent only records it. The decision then leaves the pending list.
What it helps with
- record an owner's answer to a pending task decision
- save the owner's wording alongside the selected option
- close a pending decision after its answer was sent with
agent_send
Example prompts for the agent
- Record my answer, “choose the first option,” for task decision
todo_123. - I choose “Defer.” Save my answer to this task decision.
- I already sent the answer with
agent_send; record it fortodo_123.
How to install task_decided
Manually: put the files in place
.claude/agents/task_decided.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/task_decided.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_decided.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_decided
How do I install task_decided in Claude Code?
Add .claude/agents/task_decided.md to the project.
What does task_decided do?
It records the owner's words and chosen option for a pending task decision. If the answer was already sent to the task agent with agent_send, it only saves the record.
What information does task_decided need to record an answer?
It needs the task ID, the exact label of the chosen option, and the owner's words. If the answer matches none of the options, it asks for clarification or for the decision to be revised with task_review.
Source text
task_decided.md · MIT
Record the owner's answer to a task's waiting decision (a needs_user decision on its to-do) when they gave it to you in this conversation, in their own words, instead of on the decision's card or row. Only from the main conversation or the task's supervisor, and only what the owner said: never your own call (a decision that is yours is no decision: send the agent on, or review the task). option is the label of one of the decision's options, the one their words choose; when their words fit none of them, don't record it: ask them, or restate the decision with task_review. With told_agent false the hub takes the option as the card does: it goes to the task's agent and is recorded (refused, and nothing recorded, while the agent cannot be reached). With told_agent true you already passed their answer to the agent with agent_send, and it is only recorded. Either way the decision ends as if they had answered it on its card: it leaves their Needs you, is not raised again, the to-do lists it among its decisions with their words, and the task is supervised as before. So when the owner answers a task's open decision in the chat, record it here; relaying it with agent_send alone leaves the decision waiting.
```json
{
"type": "object",
"required": ["todo_id", "option", "said"],
"properties": {
"todo_id": {
"type": "string",
"description": "the task's todo id (todo_…)"
},
"option": {
"type": "string",
"description": "the label of the option the owner chose, exactly as the decision lists it"
},
"said": {
"type": "string",
"description": "what the owner said, quoted (one or two sentences)"
},
"told_agent": {
"type": "boolean",
"description": "true when you already sent their answer to the task's agent (agent_send): it is only recorded; default false: the hub sends the option to the agent as the card does"
}
}
}
```
© nwverse · MIT · nwverse/persona