PERLY n8n Workflow Doctor
What this does
Send an exported n8n workflow (the JSON you get from Download in the n8n editor) to your own agent. You get back a severity-ordered defect report and a runnable regression test suite for its Code nodes.
Nothing is called. No credentials are used. No service is contacted. This is static analysis of the file you provide.
It runs on your machine, not on ours
You download the package once and run it inside your own agent. Your workflow file never leaves your computer — it is not uploaded to this marketplace, not sent to us, and not sent to a model provider by this skill. For a file that can contain API keys, endpoints and customer data, that is the point.
What it checks
1 · Structure — broken connections pointing at nodes that do not exist, duplicate node names (which silently break $('Node') references), missing typeVersion, malformed graphs.
2 · Leaked secrets — API keys, bearer tokens, GitHub and Slack tokens, literal passwords, personal data. The whole file is scanned: Code-node JavaScript, expressions, HTTP headers and query strings, every parameter value. Detected values are never printed back — you get the node path, the classification and the fix.
3 · Canvas geometry — sticky notes covering nodes, nodes sitting on a note's title, overlapping notes. This is measured, not eyeballed: it is also the single most common reason n8n rejects a template submission.
4 · Code-node logic — items not returned correctly, ambiguous $json under run-once-for-all-items, missing error branches, HTTP calls with no retry or timeout.
5 · Regression tests — a fixture suite for the Code nodes in your workflow, generated from your actual logic, ready to run with the MIT-licensed n8n-testkit. It fails loudly the day someone edits that logic and the output drifts.
What you get back
A single report ordered by severity, each finding with the exact node, what is wrong, why it matters, and the precise correction — plus the test suite as a file you can keep and run in CI.
Where this stops — stated up front
The tests cover the JavaScript logic in Code nodes, not the whole workflow. Your integrations are not executed, credentials are not validated, and what a connected system actually returns is not confirmed. Those must be verified in your own n8n environment before you switch anything on. A clean report means the implemented checks passed — it does not prove the workflow is correct. Better to read that here than to discover it later.
What you need
Python 3.9 or later, Node.js 18 or later, and an agent that loads SKILL.md (Claude Code and equivalents). Nothing else is installed and no account is required. Unzip the package into your skills directory and ask your agent to diagnose a workflow.
Updates and licence
Buy once. Every later version of this skill is yours at no extra cost — you are never asked to pay again for an update.
The bundled test runner n8n-testkit is ours and MIT-licensed; its licence text ships inside the package. The diagnostic scripts and rule catalogue are licensed for your own use and are not for redistribution.
Who built it
Built by PERLY — perly.dev. Two workflow templates accepted into n8n's official public library; the test runner is open source and MIT-licensed: n8n-testkit.
Questions or a problem after the download: [email protected] — answered within one working day.
