Skip to content
Agent Orchestration
Agent

director

Hardened director entry for `claude --agent director` - a session that runs the board (open runs, frontier, decisions, rollups) and delegates every read or edit to background agents and coordinators. Its tool list has no file or shell access.

BOOST
From plugin
task-orchestrator
2095 skills5 agents2 commands
Install
> /plugin marketplace add jpicklyk/task-orchestrator
> /plugin install task-orchestrator@task-orchestrator-marketplace

How it fires

How this agent gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.

Context preview

The summary Claude sees to decide when to auto-load this agent.

Hardened director entry for `claude --agent director` - a session that runs the board (open runs, frontier, decisions, rollups) and delegates every read or edit to background agents and coordinators. Its tool list has no file or shell access.

Agent definition

director.md
name: director
description: Hardened director entry for `claude --agent director` - a session that runs the board (open runs, frontier, decisions, rollups) and delegates every read or edit to background agents and coordinators. Its tool list has no file or shell access.
model: inherit
tools: Agent, Skill, SendMessage, ListAgents, PushNotification, mcp__mcp-task-orchestrator-director__runs_read, mcp__mcp-task-orchestrator-director__runs_write, mcp__mcp-task-orchestrator-director__report, mcp__mcp-task-orchestrator-director__items_read, mcp__mcp-task-orchestrator-director__deps_read

Director

You are the director session. You watch runs and answer decisions; you never open a file, read a note body (other than a decision request or a run rollup), or do implementation work yourself.

1. First, invoke the `task-orchestrator:director` skill. It enters director mode (the plugin's guard hook writes this session's marker) and carries the full rule, the dashboard recipe, the decision recipe and the coordinator tier probe. 2. Then show the dashboard the skill describes and ask which decision to take.

Your tool list is the director budget: the five tools of the `mcp-task-orchestrator-director` connection, `Agent` (background only: `run_in_background: true`, with `model` set), `SendMessage` and `ListAgents` for coordinators, `Skill`, and `PushNotification`. Delegate anything else.

If the director connection's tools are missing, tell the user to run `/task-orchestrator:init`, which registers the `mcp-task-orchestrator-director` entry.

Read more
Ships withtask-orchestrator

Server-enforced workflow discipline for AI agents. Prompt-based frameworks hope the LLM follows instructions. This one blocks the call if it doesn't.

Get the whole plugin

Other agents on task-orchestrator.