Prevent duplicate Actions
Repeated events, Action tests, automatic retries, and manual recovery can all repeat external work. Decide what one business operation means, give it a stable identity, and place duplicate checks before its write Actions.
For example, use a provider event ID to send one notification per event, or an order ID to process an order once. Those choices are different: using only the order ID will also suppress later updates to that same order.
Use Only new items
Add Only new items — Remember processed IDs and send only new items before the work to protect. Its editor supports two input types:
| Input type | Configure | Following work |
|---|---|---|
| Single event or record | Choose Record to check and a Unique ID field within that record. | Connect the next Action directly. Repeated IDs stop at this Step. |
| List of records | Choose List to check and a Unique ID field within each item. | Loop over this Step's items output and put the protected Action inside the Loop. |
Select an actual object or list through the available controls. Enter an ID field such as id or a stable individual URL, relative to the selected record. A timestamp generated on every run or a random value will not identify repeats.
Each item must have a nonempty string or finite numeric ID. Numeric and text IDs are distinct identities. The input is limited to 1,000 records and 500,000 serialized UTF-8 bytes per run. Missing IDs fail validation instead of silently inventing an identity.
Choose the first-run behavior
Check On the first run explicitly:
- Remember existing items without sending records the first observed set as the baseline. That initial set is not sent downstream.
- Send existing items too allows the initial set to be processed as well.
For a periodically fetched list, remembering the initial inventory can prevent a flood of old notifications. For a single new webhook event, silently remembering the first record may not be the intended behavior. Review the setting after changing Input type rather than relying on a default.
The first run here means the first normal execution for this Step's saved history. A preview does not establish a baseline.
Understand confirmation and scope
In single-record mode, the ID is remembered after the Workflow succeeds; overlapping runs reserve the same identity while processing. In the list pattern, an item is remembered after the protected Action succeeds. Follow the supported Loop-with-Action pattern rather than assuming arbitrary downstream work has one transaction.
History belongs to the team, Workflow, and Only new items Step. A copied or imported Workflow, or a new Step, has a separate identity and history. It does not automatically share the original's processed IDs.
Changing the selected collection, ID field, or input mode after history is initialized can conflict with that history. The runtime can require a new Only new items Step. Treat this as a new processing history and choose its first-run behavior deliberately; do not expect editing a field to migrate previous IDs.
Preview without consuming items
Select Preview sample IDs to validate the sample and count unique IDs within it. The result explicitly says saved history was not read or changed. It does not tell you whether those IDs have already been processed in normal runs.
The built-in test and Auto-validate also use a stateless preview. Auto-validate does not execute subsequent Steps on that route. To check real suppression, use a controlled normal run, repeat the same ID, then introduce a different ID and inspect both history and the target app.
Keep provider idempotency where available
Only new items reduces repeated processing; it does not make every external request exactly once. A provider may accept a write and then time out before FastHook records success. A retry can repeat that effect.
Where the Action supports a provider idempotency key, map a stable key for the intended operation. An upsert or update by a stable external identifier may also fit the business process better than an unconditional create. Use only the options actually offered by that provider Action.
A lookup before creating a record can detect an earlier result, but two simultaneous runs can both find nothing. Use the provider's atomic create, unique key, or idempotency mechanism when the operation requires it.
Check the paths around your guard
Actions before Only new items are not protected by it. Individual Test step calls can also execute Actions directly without running an upstream guard. Another Path may target the same destination independently.
Before manual recovery, check what already succeeded and how the selected recovery mode interacts with the guard. A new full run is not automatically a harmless replay. See Workflow testing, Loops, and Run history.