Skip to content

Skill Node

The Skill node is the workhorse of a DAG workflow. It runs a skill — a reusable LLM prompt — on the assigned minion, takes upstream output as input, and produces output for downstream nodes.

Use a Skill node whenever you want an LLM task to execute as part of the pipeline: research, generation, classification, code edits, etc. Multi-step work should be broken across several Skill nodes rather than packed into one large prompt — each node is independently retriable and visible in the run timeline.

FieldDescription
skill_idThe skill to execute. Pick from the workspace’s skill library.
skill_version_idWhich version of that skill to pin. Defaults to the current version. Pinning an older version protects running workflows from skill edits.
assigned_role"pm" or "engineer" — which role in the project is responsible for this step.

Reviews are not configured on the Skill node itself. To require approval after a skill, add a separate Review node after it.

  1. When cascade reaches the node, it is inserted as pending and queued for the minion assigned to the project for this role.
  2. The minion picks up the node, runs the skill with the upstream output as input, and reports success (with output) or failure.
  3. On success, cascade fires to downstream nodes. On failure, the graph halts at this node unless a downstream recovery path exists.
  • skill_id is required.
  • skill_version_id is required.
  • assigned_role is required.

If the skill referenced by skill_id is deleted from the workspace, the validator flags the node as broken.