Runs

Testing before activation

The Inspector's Test tab: run steps against real data, pin sample outputs, capture real webhook/email/text samples, and preview sends without sending.

Everything can be tested before a workflow is live — that's what the Inspector's Test tab and the step pins are for.

Pins: each step's sample data

Every step has a pinned output — a sample of what it produces, used as the next step's test input. Pins are what make the chain testable: step 3 runs against step 2's pin, not against production.

The Inspector Test tab with a pinned output Run step executes for real against the upstream pin; the result becomes this step's pin.

  • Run step executes the step for real (a web request actually fetches) against the upstream pin, and pins the result.
  • Paste fixture / Edit JSON — pin data by hand when you'd rather not run anything.
  • If you change a step's config after pinning, the pin is flagged stale — run again to refresh.

Capturing real trigger samples

Trigger pins are captured on the Config tab, from the real source — because guessing a webhook's payload shape is how workflows break on day one:

  • Webhook — capture sample arms a listener; send a test event from the sender and its payload becomes the sample. Nothing runs — capture only records the shape.
  • Email / text triggers — test with a real email / text: send one, the sample is captured.
  • File upload and form triggers capture from a real upload/submission the same way.

Sends preview, they don't send

Side-effect steps (send email, send text, connected-app writes, non-GET web requests) never fire in test mode. Instead, Preview request renders the exact outbound payload — rendered against real upstream data — with the note "nothing was sent." What you see is byte-for-byte what a live run would send.

Two more safety rails:

  • Testing a loop runs the body once with a sample item.
  • Call workflow simulates in test — it doesn't fire the target.

The verify walk

When you build with AI, all of this is driven for you: the walk runs each step, pins real outputs, previews sends, and marks plan rows verified — pausing only for things only you can provide (a connection, a real sample, a decision). Details in how verification works.

Activation stays human-only: when every row is verified, the topbar live switch is yours to flip.

‹ Activation & snapshots Error handling ›