Creating a test set
Create a test set that validates the agent's inputs, outputs and expected behaviour.
Creating a test set
Click + Create set on the Test sets page to open the creation panel. The same panel edits existing sets — the title then reads Edit test set.
Identification
Name
Required. Give it a descriptive name that says which scenario is being tested.
- For example: "Happy path", "Connection error between nodes", "Invalid input with no configuration"
Expected result
Whether the agent should succeed or fail in this scenario. Click the card to switch between:
| Option | What is expected |
|---|---|
| Should succeed | The agent should complete the run without escalating |
| Should fail | The agent should escalate or reject this input |
Input arguments
Lists every input argument configured on the agent. For each one:
- Enter the value that will be sent to the agent in this test.
- Click Skip to leave the argument out of this evaluation — its value is not checked.
The input control depends on the argument's type:
| Type | Control |
|---|---|
| Text | A free text field |
| Number | A numeric field |
| Date | A text field in YYYY-MM-DD format |
| Boolean | A True / False toggle |
| File | A Select file button (or Replace / Remove when a file is already there) |
| JSON | A text editor in { "key": "value" } format |
If the agent has no input arguments configured, the section says "This agent has no input arguments configured."
Expected outputs
Lists every output argument configured on the agent. Enter the values the agent must return for the test to pass.
As with the inputs, click Skip to leave a particular output argument unchecked.
If the agent has no output arguments configured, the section says "This agent has no outputs configured."
Saving
Click Create set (or Save changes when editing) to confirm. The set then appears in the list on the Test sets page.
Use clear names and varied scenarios — test the happy path, invalid inputs and edge cases. That is what catches a regression before a new version ships.