Automation repair · n8n, webhooks, scripts
A broken automation, fixed with a test you can run.
For small teams and agencies whose workflow has stopped working. We reproduce the failure from your public workflow first. If we can’t, you pay nothing. Then you get a fixed price, the fix, an acceptance test that passes on your side and a rollback to your current version.
Opens an email to hello@cybernative.ai. Send the link to your public workflow, repository or issue and the output you expect. Do not attach files or private records.
See a repair proof- No reproduction, no charge
- Fixed price before any payment
- Deposit refunded if the test can’t pass
| The record that fails today | the correct result |
|---|---|
| A record that worked before | the same result as before |
| A bad or rejected record | an alert, never a false success |
| The same records, run again | no duplicates |
| Roll back to your version | your old behavior returns |
Repair proof · lead cutover
9 of 9 checks pass on a rebuilt Mailchimp → Brevo cutover.
For a repair, we reproduce the failure on sample data before quoting. For a new flow, move or extension, we check feasibility against your inputs and expected outputs. This example rebuilds a publicly requested migration. It was not a customer job: three existing website forms, one welcome email, no duplicates.
| # | Situation | Observed result |
|---|---|---|
| 01 | Before the cutover | PASS One popup lead reaches Mailchimp only and gets its welcome. Brevo: 0. |
| 02 | Same person, four spellings, three forms | PASS 1 Brevo contact, 1 welcome, 0 Mailchimp calls. |
| 03 | Invalid email address | PASS Alert raised; no contact created. |
| 04 | Brevo rejects the lead (400) | PASS Alert raised; no contact created. |
| 05 | Brevo rate limit (429) | PASS Retried once after a pause; contact and welcome created once. |
| 06 | Brevo says OK but saves nothing | PASS The check reads the list back, fails, and alerts. |
| 07 | Old Mailchimp route left on | PASS Two welcomes detected and flagged. Route off: one welcome, no alert. |
| 08 | Roll back to the old version | PASS The next new lead goes to Mailchimp again. |
| 09 | Vendor failures on demand | PASS The test double replays 401, 429, a renamed field and a second page. |
| — | Can the checks fail? | With the duplicate-welcome detector removed, check 07 fails. The check is doing the work. |
9 / 9 pass
Each check makes one real situation happen and records what the workflow did. A check that cannot fail proves nothing, so the last row removes a detector and shows its check failing.
This is a rebuild, not the requester’s own scenario, accounts or mail. An alert here flags a failed lead but does not store it for replay; on a real job that is part of the hardening scope.
How a repair runs · five steps
We check the job before you pay.
The clock starts only when the scope and your examples are agreed. Anything outside the written rows is quoted separately before we do it.
You send a public link
A public repository, issue, forum thread or shared workflow with credentials removed, and 5–10 made-up example records with their expected outputs, posted there. For a repair, include a failing example. For a new flow, move or extension, no failure is required.
We check the samples
For a repair, we reproduce the failure; if we can’t, you pay nothing. For a new flow, move or extension, we check feasibility against your expected outputs. No deposit is due before scope and a feasible acceptance test are agreed.
You accept a fixed price and the rows
A written scope, the acceptance rows and the number of business days. You pay 50% when you accept them.
We hand over the fix
The working change, an error path with an alert, your previous version saved for rollback and a one-page handover. You apply it on your side with the apply script we include, and run the acceptance test there.
You run the rows
The other 50% is due when they pass on your side. If we cannot make them pass within the agreed days, the deposit is refunded in full.
Scopes · one fixed price per job
Three sizes of repair.
We confirm the scope after reproducing the failure or checking feasibility for a new flow, move or extension, then quote one fixed price for it. If the platform’s own feature can replace the part that broke, we quote that smaller job instead, or tell you how to do it yourself.
- Scope 1
Fix
One failing step or connection: an expired login, a field renamed by an app update, pagination, a changed payload, a missing retry.
Acceptance rows on up to 10 of your records, an alert instead of a silent failure, rollback and a handover note.
- Scope 2
Fix + harden
The fix, plus the whole workflow (about 20 steps, up to 3 connected apps): no duplicates on re-run, retry with back-off and an error queue.
Replay of the records that failed during the outage, which you run on your side from our handover, where the source or the provider still holds them. We tell you in writing which ones cannot be recovered.
- Scope 3
Fix + harden + custom step
Both of the above, plus one custom code or API step, such as an OAuth refresh, cursor pagination or signature check, with unit tests.
A 7-day watch after handover, with one fix for the same kind of failure.
Each job runs on a written scope and acceptance sheet you confirm before paying. A failing row you report within 5 business days of handover gets one fix cycle; the Fix and Fix + harden scopes include 7 days of support for the same failure.
Fit · before you write
If a free answer fixes it, use it.
We are worth paying when the free answers haven’t worked, when runs were retried before the fix, when one run handled several records and only some went through, when runs show success but did nothing, or when you need proof that it is fixed.
Free first · n8n
Fix the cause, then press Retry once on each failed run.
If the failed runs are still listed under Executions and nobody has retried them, choose Retry with currently saved workflow on each one, so the retry uses your fix. In our tests that recovered each record once. You don’t need us for that. n8n deletes old runs (Cloud Starter after 7 days, Pro after 30, self-hosted after 14 by default, and sooner on a busy instance), so do it soon.
Don’t press Retry before the fix. A retry that fails becomes a second run of the same record, and retrying both sends it twice.
We take
- n8n workflows, webhook and HTTP API integrations, and scripts in Python, JavaScript or Google Apps Script.
- Zapier or Make, when the workflow and sample payloads are public.
- A named failure in a workflow that used to work, or was half built, where “fixed” can be written as input → expected output.
- A small new flow, a move between tools or an extension of a working one: up to about 20 steps and 3 connected apps, with “done” written as input → expected output. With nothing to reproduce, we check your samples and expected outputs before any payment.
We decline
- Health, payment-card or government-ID data.
- Writes to a production database, or access to your systems or credentials of any kind.
- An AI-agent product that takes open-ended actions, “make it smarter”, or holding or bulk-moving your customer records ourselves.
- A failure we cannot reproduce from what you can share.
Start with a public link
Send the link and the result you need.
Post the workflow with credentials removed, plus made-up example records and the output each should give, somewhere public. For a repair, include one that fails. For a new flow, move or extension, no failure is required.
Email us the linkOpens an email to hello@cybernative.ai. Send links, not attachments. Never post or send passwords, API keys or real customer records.