Route n8n workflow tasks to specialist guidance skills

This skill is a router, not a reference - it tells an agent which specialist n8n-mcp skill owns the rules for whatever it's about to do, invoked at the start of any n8n MCP workflow design, inspection, edit, validation, test, deployment, credential, execution, or troubleshooting task. It starts with read-only discovery and live schema inspection, never copies secrets into prompts or workflow fields, never infers the target instance, and requires approval before tests with side effects, activation, deletion, credential mutation, or other externally visible changes. Because the community n8n-mcp server and n8n itself move faster than any model's training cutoff, when a skill's described tool, parameter shape, or behavior drifts from what the live tool actually shows, the live tool wins - flag the drift and suggest updating the pack and instance.

Three non-negotiable rules, each preventing a workflow class that looks correct but breaks in production: invoke the relevant skill before any n8n action, not just MCP calls - PreToolUse hooks remind on the highest-impact calls only when the plugin bundle is installed, so on plain skill-upload platforms like Claude.ai the responsibility is entirely the agent's own; validate AND verify before activating, since validate_workflow only proves well-formed JSON, not correctness - always also call n8n_get_workflow after every create or update to inspect the connections object, since validation alone misses silently dropped wires, Merge index off-by-ones, and unwired error outputs; and secrets never go in text fields - tokens, API keys, and passwords go through the n8n credential system, falling back to the HTTP Request node with the official credential type if no native node exists, since a Set node holding a token referenced via an expression is a leak with extra steps.

Because n8n changes constantly and "remembered" parameter names often validate as plain strings while doing nothing at runtime, the skill instructs trusting live tools (get_node, search_nodes, tools_documentation) and skill bodies over recollection - if a skill contradicts memory, trust the skill; if get_node contradicts a skill, trust the tool and flag the drift. Four strong defaults (each skill can carve its own exceptions): the Code node is a last resort behind an expression and then an Edit Fields arrow function; a Set node feeding zero or one consumers is almost always wrong and should be inlined at the consumer instead; per-item iteration is automatic, so a Loop Over Items node to "make it loop" is usually unnecessary; and nodes should always be configured from the live schema via get_node, never from memory.

Fifteen "red flag" thoughts each map to a skill to invoke first: assuming a workflow is simple enough to just build (most ship at 10+ nodes) routes to n8n-workflow-patterns; reaching for a Set node to map fields routes to n8n-expression-syntax; defaulting to a Code node routes to n8n-code-javascript; writing Python without an explicit ask still routes to n8n-code-javascript (Python only on request, via n8n-code-python); writing code an AI agent will call routes to n8n-code-tool (a different runtime contract); reaching for a DateTime node for date math routes to n8n-expression-syntax (inline Luxon is usually right); wiring a three-source Merge routes to n8n-node-configuration (Merge defaults to two inputs, the third silently drops); treating a passed validation as activation-ready, or hitting a validation error, both route to n8n-validation-expert; referencing $json.x directly in a branchy workflow routes to n8n-expression-syntax (prefer $('Node').item.json.x); building a happy-path-only webhook or scheduled flow routes to n8n-error-handling; passing a file through as JSON routes to n8n-binary-and-data (file contents live in $binary and can't cross the agent-tool boundary); wiring an AI agent with tools routes to n8n-agents (tool names and descriptions are the prompt); copying logic into another workflow or noticing a workflow getting big routes to n8n-subworkflows; and creating a credential or opening a workflow on a multi-instance account routes to n8n-multi-instance (every call hits the currently-targeted instance, and an ambiguous credential write fails closed).

The full skill index spans fourteen entries covering tool/node discovery and credentials (n8n-mcp-tools-expert), workflow architecture (n8n-workflow-patterns), node configuration (n8n-node-configuration), expressions and data mapping (n8n-expression-syntax), validation interpretation (n8n-validation-expert), JavaScript and Python Code nodes (n8n-code-javascript, n8n-code-python), the Custom Code Tool contract (n8n-code-tool), error handling (n8n-error-handling), binary/file data (n8n-binary-and-data), sub-workflow extraction (n8n-subworkflows), AI agents (n8n-agents), and multi-instance targeting (n8n-multi-instance).

The skill also gives working knowledge of the n8n-mcp tool surface itself, grouped by purpose: discovery/docs (tools_documentation, search_nodes, get_node, validate_node, search_templates/get_template), build/edit (n8n_create_workflow, n8n_update_partial_workflow as the preferred incremental-diff editor, n8n_update_full_workflow, n8n_autofix_workflow, n8n_deploy_template), validation (validate_workflow taking long-form node types, n8n_validate_workflow by id), inspect/lifecycle (n8n_get_workflow, n8n_list_workflows, n8n_delete_workflow, n8n_workflow_versions, n8n_instances, n8n_health_check), test/run (n8n_test_workflow which fires real side effects, n8n_executions, and n8n_evaluations for n8n 2.30+ test runs), and data/credentials/audit (n8n_manage_datatable, n8n_manage_credentials, n8n_audit_instance) - flagging a node-type form trap where get_node/validate_node take short form while workflow JSON uses long form. The overall protocol: recognize the matching skill and invoke it before the first MCP call, skim tools_documentation once per session, call get_node before configuring any node, validate before activating and get_workflow after to check connections, and surface any drift noticed. When nothing fits cleanly - a workflow the user built in the UI can't be found (usually per-workflow MCP access is off), the user reports something broken (believe them and re-check against get_node), or no skill fits a non-trivial task - the guidance is to ask rather than guess, since these are opinionated best practices, not laws, editable as plain markdown.

Use this router at the start of any n8n MCP workflow design, inspection, edit, validation, test, deployment, credential, execution, or troubleshooting task, invoking the specialist skill it names before the first MCP call. Never copy secrets into prompts or workflow fields, never infer the target instance, and obtain approval before any test with side effects, activation, deletion, credential mutation, or other externally visible change.

Input is an n8n MCP task description. Output is the identified specialist skill(s) to invoke, plus, when actually building, a validated and verified workflow whose connections were confirmed post-edit via n8n_get_workflow.

request: Build a webhook that validates input, calls an API, and returns structured errors.
specialists: [n8n-workflow-patterns, n8n-node-configuration, n8n-error-handling]
sequence:
  - inspect the target instance and live node schemas
  - build and validate the graph
  - preview side effects and obtain approval
  - write changes, fetch the saved workflow with n8n_get_workflow, and revalidate
  - activate and test only after approval

Every other n8n-mcp skill in the family - n8n-mcp-tools-expert, n8n-workflow-patterns, n8n-node-configuration, n8n-expression-syntax, n8n-validation-expert, n8n-code-javascript, n8n-code-python, n8n-code-tool, n8n-error-handling, n8n-binary-and-data, n8n-subworkflows, n8n-agents, and n8n-multi-instance - plus the full n8n-mcp tool surface (discovery, build/edit, validation, lifecycle, test/run, and data/credential/audit tools).

Coding agents and n8n workflow builders using the n8n-mcp server who need to know which specialist skill governs a given step, rather than guessing at parameter names or tool behavior from training data.