Workflows

Workflows define multi-step agent pipelines — sequences of phases that transform a task from specification to deployed code.

Workflow definition

Workflows are markdown files with a structured format:

---
name: Widget Pipeline
description: Build, test, and deploy widgets end-to-end.
status: active
---

## Steps

### research
- skill: researching-codebase
- timeout_minutes: 15

### plan
- skill: planning-implementation
- timeout_minutes: 20
- depends_on: research

### build
- skill: implementing-code
- timeout_minutes: 60
- depends_on: plan

### test
- skill: iterating-on-tests
- timeout_minutes: 30
- depends_on: build
- gate: true
- max_retries: 2

### review
- skill: packaging-for-review
- timeout_minutes: 15
- depends_on: test

Workflow fields

Field	Type	Description
name	string	Display name.
description	string	What the workflow does.
status	string	draft, active, or archived.

Step fields

Field	Type	Description
skill	string	The skill ID to execute for this step.
timeout_minutes	number	Maximum time before the watchdog kills the step.
depends_on	string	Step ID that must complete first.
gate	boolean	If true, step must pass before successors run.
max_retries	number	How many times to retry on failure.

Running workflows

Workflows can be started through:

1. MCP tool: session_delegate({ workflow_id: "widget-pipeline", project: "/path/to/project" })
2. Dashboard: Click “Run” on a workflow in the Workflows view
3. Automations: Triggered on a schedule or event

Note

Each workflow run creates an isolated session. The session progresses through steps automatically, using the skill defined for each step.

What’s next

• Custom UI — give your plugin its own views, panels, and command-palette entries in the workspace.