Workflows

Workflows let you define multi-step processes where tasks flow between agents automatically. A workflow is a DAG (directed acyclic graph) of steps — each step creates a real Mission Control task assigned to an agent, and when that task completes, the engine advances to the next step(s).

Key Concepts

Steps

Each workflow is made up of steps. A step is a single unit of work — it has a name, an assigned agent, an action prompt, and optional dependencies on other steps.

DAG (Directed Acyclic Graph)

Steps are connected in a DAG — meaning work flows in one direction without cycles. Steps with no dependencies (root steps) run first. Steps that depend on other steps wait until all their dependencies complete.

Executions

When you run a workflow, it creates an execution — a live instance that tracks the progress of every step. You can watch executions in real-time, cancel them, or retry failed ones.

Task Integration

Each workflow step creates a real MC task. This means all existing task features (comments, @mentions, agent memory, approvals) work inside workflows.

Plan Requirements

Feature	Plan
Create workflows	Pro+
Run workflows	Pro+
Install from marketplace	Pro+
Unlimited workflows	Scale+

Step Types

Agent Steps

Assigned to an AI agent. When the step starts, a task is created and the agent processes it automatically (if in auto mode) or waits for manual execution.

Human Review Steps

No agent assigned. Creates a task that a human must pick up and complete. Useful for approval gates and manual quality checks.

Approval Steps

Steps with "requires approval" enabled. The workflow pauses at this step until a human approves the task.

Triggers

Currently, workflows support manual triggers — you click "Run" to start an execution. Future versions will add scheduled (cron) and event-based triggers.