n8n Sub-workflows is a skill for Claude that treats a sub-workflow as a reusable function. An Execute Workflow Trigger declares typed inputs, the body does the work and the last node returns the output, which a caller invokes through an Execute Workflow node.
The skill covers when to extract logic, searching for an existing sub-workflow first, typed Define Below inputs, return contracts, the mode and waitForSubWorkflow settings, splitting by input shape and using a typed sub-workflow as an AI agent tool.
What You Get
- A decision flow for whether a chunk of logic should become a sub-workflow
- Rules for typed workflowInputs on the Execute Workflow Trigger, with the two exceptions for binary and zero inputs
- Contract rules: describe inputs and outputs, return natural shapes, return ok and error for expected failures, freeze fields once callers exist
- A table comparing mode all and mode each, and the waitForSubWorkflow setting
- The N+1 pattern for sub-workflows with different input shapes
- Reference files for naming and discovery and for sub-workflow patterns
Common Use Cases
- Extracting parsing, retry or notification logic that several workflows repeat
- Breaking up a workflow that has grown past ten or more nodes
- Exposing a multi-step operation as an AI agent tool
- Running many items in parallel with fire-and-forget calls and a Data Table to track them
- Renaming a sub-workflow input without breaking its callers
What Makes It Unique
It is built around n8n’s limits: the MCP cannot filter workflows by tag, so naming is the search surface, and n8n does not error on an unknown input field, so a rename can break callers silently. It also explains that mode each with waitForSubWorkflow set to false is the only real parallelism, and it tells Claude to search the library before writing new logic.
Under The Hood
The skill is instructions and two reference markdown files, with no scripts. It tells Claude to use the n8n-mcp tools to list and read workflows, patch nodes and connections, validate, run isolated test executions with pinned data, read executions, and manage Data Tables, all against your n8n server. It can create and activate workflows there, and a fire-and-forget design makes sub-workflows run in the background.