Subagents
Three purpose-built agents, each specialized for one phase of the workflow lifecycle.
n8n-workflow-builder
Invoke when: starting a new workflow from scratch.
Orchestrates the full build loop: discovers nodes, pulls schemas, validates configs, creates the workflow, validates end-to-end, activates.
Delegates to:
n8n-workflow-validatorwhen validation gets complexn8n-workflow-deployeronce the workflow exists and needs iteration
Won't:
- Hand-write node configs from memory
- Try to build in one shot (always iterates via
n8n_update_partial_workflow)
n8n-workflow-validator
Invoke when: you have an existing workflow (by ID or JSON) that needs checking.
Runs validate_workflow with the appropriate profile, interprets errors in plain language, and proposes fixes. Can optionally run n8n_autofix_workflow — always previews changes first.
Profile selection:
runtimeby default (covers values + types)strictbefore production hand-offminimalwhen debugging connection structureai-friendlywhen validating AI-generated configs
Delegates to:
n8n-workflow-deployerto apply proposed fixesn8n-workflow-builderif the workflow needs to be rebuilt from scratch
n8n-workflow-deployer
Invoke when: modifying, activating, or inspecting a workflow that already exists.
Handles:
- Partial updates via
n8n_update_partial_workflow(withintent) - Activation / deactivation
- Execution diagnosis via
n8n_executions
Won't:
- Use
n8n_update_full_workflowwhen a partial update would do - Skip the
intentparameter - Activate a workflow with validation errors
How they fit together
user: "build me a webhook → AI → Slack workflow"
│
▼
n8n-workflow-builder
│
├── (validation gets weird) ──▶ n8n-workflow-validator
│
├── (workflow created, now iterate) ──▶ n8n-workflow-deployer
│
└── (activate) ──▶ doneAll three reference the bundled skills — n8n-mcp-tools-expert, n8n-workflow-patterns, n8n-node-configuration, n8n-validation-expert. You don't need to invoke the skills manually; the agents do.