XIO

xWire

Posts from every feed QUEST has found across the web — blogs, news, podcasts, channels and more.

n8n Community - Latest posts

Latest posts

Why my approval webhook kept leaving people with a spinning browser (Sharing my learnings)

Thanks for sharing this, Felix. Mahdi’s GET/POST split covers the scanner side well. One case none of the replies mention: the approval nobody ever clicks. The approver is on holiday, the email lands in spam, or it gets buried. The workflow sits in its Wait node for days. Then someone approves a week later, and the refund or quote goes out after the customer was already handled another way. What I do now: Store an expiry time on the approval record when you create it (24h works for most of…

Regression in 2.46.0: $('Node').first() returns null for a node's error output in webhook executions (fixed by N8N_EXPRESSION_ENGINE=legacy)

Your table already shows the split: $json and the node right after the error output are fine, and only $(‘Thrower’) goes null. The try/catch line is the useful clue, it dies with “could not be cloned” instead of returning the item. That matches the expression engine added in 2.46 evaluating production webhook runs in a separate isolate. The error output item still carries a real Error object, and that object cannot be cloned across the isolate, so the lookup comes back as null and n8n does not…

Help - all my workflows are down

I meant the app that normally sends items to n8n. For example, if a form starts the workflow, does the form itself show any submissions after the last automatic run? If it does, you can compare those with the ones your manual runs picked up. If the source has no history, the execution list alone can’t tell you what was missed. Thanks for clarifying, Shannon.

Reasoning_effort error with n8n Assistant and OpenAI-compatible endpoint

Yes, you’re right that this works. I found the reason for my issue in the meantime: I used the UI to set up Assistant (stupid me, using the convenient way!) and with a little help of the AI digging into the 2.48.5 source, I found that if you set up Assistant through the UI with a custom OpenAI-compatible endpoint, you need to set N8N_INSTANCE_AI_REASONING_EFFORT=none because (written by AI, not me): When you configure an OpenAI credential with a custom URL through the Assistant UI, n8n assigns…

Would you let an AI agent fix your production n8n workflows without your approval?

@elmirbesirovic you’re right, and that trap caught me once. The duplicated workflow pointed at the same credentials and the first ‘test’ run emailed real customers. Now I keep sandbox credentials in a separate project and the agent only gets access to that one. If a fix can’t be tested against sandbox data, that’s usually a sign it’s touching prod too closely anyway. The diff hash is a good control. The weak point stays human: long diff, tired person, approve wins. So I keep the approval list…

Would you let an AI agent fix your production n8n workflows without your approval?

Kartike’s decay point matches what I’ve seen. The fix that worked best for me is asking less often, not more carefully. If every change needs a click, the click becomes a habit within weeks. If the rule is “only ask when it touches a CRM write, an email or money above X”, approvals stay rare, so people actually read them. The second thing that helps is showing what changed next to the request, not just “approve the fix?”. A click on “these 2 fields now write to a different column” means…

Postgres connections with custom SSL certs finally work properly (3.1)

Most of my client instances run on managed Postgres (RDS and a couple of self-hosted boxes), and until recently the Postgres credentials in n8n only handled the simple cases. If the database required a custom CA or client certs, the workarounds were ugly: relax security, or route everything through an SSH tunnel and hope. n8n 3.1 (Nov 3) fixed this properly. Postgres credentials now take a CA certificate, a client certificate, and a client private key. The fields show up when SSL mode is set to…

Would you let an AI agent modify your production n8n workflows without approval?

A fixed core that never shrinks, plus real incidents promoted after scrubbing. It also makes the fixture set an actual record of what has gone wrong. Schema drift is the part I’d worry about with full replay. One thing that could help is tagging each fixture with the upstream schema version it came from, so stale ones get flagged instead of failing silently. When an upstream schema changes on purpose, do you migrate the affected fixtures or retire and replace them?

Help - all my workflows are down

Manual trigger working while nothing else fires is usually one of three things, and none of them throw an error. First, on recent n8n versions the active toggle alone is not enough, the workflow has to be published too (that is the change the docs link above covers). Open one of the dead workflows and check its publish state, not just the active switch. Second, if it is published, open the executions list and filter to queued and running. If executions pile up there, the workflow is fine and…

Help - all my workflows are down

Hey @Shannon_Lewis, while you wait for a response, here are some things that might help: Suggested resources (click for more details) @kjooleng, @Anshul_Namdev, @Gallo_AIA - you’ve helped with similar issues before, can you take a look? Automatically suggested by n8n’s community bot. It’s a pilot - please share feedback here.

Help - all my workflows are down

Describe the problem/error/question None of my workflows are triggering. The AI said to have the owner restart the instance, which I did and it didn’t help. I can’t get anyone in support to actually support their product. No cap has been reached. Manual trigger works. What is the error message (if any)? Please share your workflow No errors. They just aren't triggering. Share the output returned by the last node Information on your n8n setup n8n version: Database (default: SQLite): n8n…

Your client noticed the automation broke before you did. Again

The error trigger and the heartbeat cover the case where the run itself breaks. The one that got me a few times is the run that succeeds and does the wrong thing. The API returns 200 with an error message inside the payload, the filter lets the wrong items through, everything is green and nothing fires. What helped us: check the outcome, not just the run. Near the end of the workflow, count how many items made it through the last step and compare that with what came in. If the numbers look…

Why my approval webhook kept leaving people with a spinning browser (Sharing my learnings)

For the email scanner issue, I’d separate opening the link from recording the decision A GET request should only display the approval page. The actual approve/reject action should happen through an explicit POST, with the decision stored against a unique approval ID. That way, security scanners can’t approve something just by following the URL, and repeated clicks can return “already processed” without executing the action again. It does add a confirmation step, but that’s a much safer…

Would you let an AI agent fix your production n8n workflows without your approval?

Fair point, this is the gap in the test-on-a-copy advice. Duplicating a workflow in n8n copies the credential IDs along with it, so the copy runs on the live CRM creds from the very first test run. What works in practice: duplicate the credentials too and point the copy at the test set, a sandbox account or a read-only key. Naming them with a TEST_ prefix makes it obvious when a copy is still holding prod creds. And for nodes that can’t be sandboxed, like the real mailbox, I keep them disabled…