Skip to main content
subagent CX Codex

ask_escalate — Claude Code subagent: what it does and how to install

The ask_escalate Claude Code subagent routes decisions that task agents cannot make to the user. It groups unanswered task questions into one reminder.

MIT ★ 395 updated 2026-11-16

Install into Claude Code · Manually: put the files in place All methods and platforms ↓
.claude/agents/ask_escalate.md
Open on GitHub

What ask_escalate does

ask_escalate sends the user decisions that task agents cannot settle on their own. It acts on an escalation message, such as a request to approve tool access or a plan, or a question about an irreversible action, a change visible to others, or a shift in the task goal.

It combines every task question the user has not yet seen into one reminder, with the reason and a recommendation. Replies go directly to the agents and are saved as task decisions. In low-distraction mode, a phone push is limited to certain ownership reasons or decisions the user must handle now.

Only the task supervisor and the main conversation can call it. A supervisor's escalation also reaches the main conversation, preventing a duplicate question to the user.

What it helps with

  • route a tool permission decision to the user
  • request approval for a task plan
  • escalate an irreversible or public choice
  • bundle pending task questions into one reminder

Example prompts for the agent

  • Ask the user to decide whether this task can use the tool, and include a recommendation.
  • Send the plan to the user for approval and save their answer as a task decision.
  • Escalate the choice that would change the goal or affect other people.

How to install ask_escalate

Manually: put the files in place

.claude/agents/ask_escalate.md

For all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention

Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean

Questions about ask_escalate

How do I install ask_escalate in Claude Code?

Add the subagent file to the project at .claude/agents/ask_escalate.md.

What does ask_escalate do?

It routes decisions beyond the task agents' authority to the user, then passes their replies back to the agents as task decisions.

Who can invoke ask_escalate?

Only the task supervisor and the main conversation. A supervisor's escalation is also sent to the main conversation so it does not ask the user twice.

Source text

ask_escalate.md · MIT
Hand the cards a task's agents wait on to the user, when a 督办 message gave them to you first and the user must decide: a tool permission or a plan (no conversation approves those), or a question that is irreversible, seen by other people, changes the goal or goes against a decision already made on the task. Every card of the task the user has not been told of goes to them at once, as one reminder (in the user's low-distraction mode it pushes only when why_owner or p0 says it is theirs alone), with your reason and recommendation on it; their answers go straight to the agents and are kept as the task's decisions. Only the task's supervisor and the main conversation can. A leader's escalation is told to the main conversation too, so don't ask the user the same thing again yourself.

```json
{
  "type": "object",
  "required": ["todo_id", "reason"],
  "properties": {
    "todo_id": {
      "type": "string",
      "description": "the task's todo id (todo_…), from the 督办 message"
    },
    "reason": {
      "type": "string",
      "description": "in a sentence, in the user's language: what they are to decide and why it is theirs"
    },
    "recommend": {
      "type": "string",
      "description": "the option you recommend, in a line (an option's label when there is one)"
    },
    "why_owner": {
      "type": "string",
      "enum": ["irreversible", "visible_to_others", "priority", "acceptance", "only_owner_knows"],
      "description": "why only the user can decide it. In the user's low-distraction mode only irreversible, visible_to_others and only_owner_knows push to their phone; the rest wait in their Needs you list and briefing"
    },
    "p0": {
      "type": "boolean",
      "description": "true: the user must do or decide this themselves now, so it pushes even in low-distraction mode whatever why_owner says"
    }
  }
}
```

© nwverse · MIT · nwverse/persona

Similar subagents

All subagents for Agent setup: memory, context, orchestration →

From the same set61

The whole set →
task_handoff The task_handoff Claude Code subagent sends part of an assigned task to a separate session through Persona. The handoff stays attached to the same to-do and supervisor. CX Codex ★ 395repo stars leader_progress The leader_progress Claude Code subagent adds direction updates to a work log. It reports completed goals and scheduled check-ins, rather than every intermediate step. CX Codex ★ 395repo stars interaction_answer The interaction_answer Claude Code subagent answers another agent’s question on the user’s behalf when the user has chosen an answer or delegated the choice. It picks from the offered options and leaves personal decisions to the user. CX Codex ★ 395repo stars agent_autoname_receipt The agent_autoname_receipt Claude Code subagent reads existing name-confirmation cards and one coding agent’s current naming snapshot. It helps inspect the available evidence about a rename. CX Codex ★ 395repo stars agent_supervise The agent_supervise Claude Code subagent takes over an agent that was started without a task and tracks its work as a task in the current conversation. It routes progress messages to a supervisor and closes the task with task_review. CX Codex ★ 395repo stars task_decided 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. CX Codex ★ 395repo stars agents_list agents_list — a subagent for Claude Code and Codex monitoring running herdr sessions across user machines. Displays active jobs, pending approvals, and subscription limits. CX Codex ★ 395repo stars task_scope task_scope — a subagent for Claude Code revises a supervised task after it starts. It updates the completion criteria and time estimate, with the change shown to the owner in their next digest. CX Codex ★ 395repo stars

Project sponsors

Companies that keep this analytics open