Tests one stage of the automation on one input.
Queues a sandbox run of the stage's agent, on its deployed revision, with exactly the request a pass would send, and returns at once; poll the test until it finishes. The input is a real item (only read), a record the source returned, typed data, the step before in a walk, or a run a pass dispatched (`executionRecordId`): a replay of that run, at its stage unless `stageIndex` names another, on exactly the data it was handed. Each connection reads for real and writes to the simulator unless routed to the simulator entirely; a live connection is not offered. The run is charged to the agent's test budget. To test a revision before deploying it, name it with `agentRevision`, or per agent with `agentRevisions` for a walk; to test an edit before it is even a revision, name the caller's draft with `agentDraft`. The test then runs that revision or draft, holds it to every check the deployed one would face, and reports `agentRevisionDeployed: false`. Returns 400 when the stage cannot be tested yet — not a PROCESS stage, a batch stage, an undeployed agent, or an agent holding a tool that would reach a real person — or the run to replay is not one of this automation's; 404 when a named revision or run does not exist; and 409 when the draft has moved past the version named, or the caller already has the most tests running at once.
Authorization
Bearer JWT Authorization header using the Bearer scheme. Enter 'Bearer' [space] and then your token.
In: header
Path Parameters
The trigger identifier.
uuidThe tenant identifier
The stage, the input, and how to reach each connection.
The stage to run, by its position in the shape — a stage being edited may have no id yet. Required unless the seed continues from a previous step, which defaults to the stage that step advanced into.
int32The shape to test, when it is a draft the editor has not saved; the automation's saved shape otherwise.
Where the stage's input comes from. Exactly one source.
For a real item only: remove what the tested stage and every later stage declare they produce, and what the item's history records them writing, so a stage re-tested on an item that has moved past it does not read its own earlier answer. On by default.
How to reach each of the stage agent's connections, by connection id: SIMULATED or
READ_THROUGH. A connection not named here reads for real and writes to the simulator, unless it already
has a sandbox of its own saved. A live connection is never offered: a write a test makes is not a ledger
effect, so the real run would make it again.
Keep what the test writes to the simulator. Off by default: a test, and every step of its walk, writes into a private copy of each simulated world, made from the shared one the first time the walk touches it, so a walk's later steps read its earlier steps' writes and no other test does. The copy is discarded when the walk's tests expire. On, the test writes into the shared world, where every later test and sandbox run reads it.
The walk to add this test to, when it re-runs a step of one rather than continuing it.
uuidWhich step of CreateLedgerStageTestRequest.WalkId this test re-runs. Requires CreateLedgerStageTestRequest.WalkId.
int32At an ASK stage, the reply a person would send — prose or an {"answers":{...}} payload — read the way
a real reply is read. Required there, and refused at any other stage.
The revision of the stage's agent to run instead of its deployed one — a revision created and not yet deployed, tested before it ships. It is held to everything the deployed one would be: the tools that would reach a person, the connections it may be routed to. Refused at an ASK stage, which runs no agent, and together with CreateLedgerStageTestRequest.AgentRevisions.
int64The revision to run per agent, by agent id, for a walk whose stages run different agents: a stage runs the revision its agent is named with here, and the deployed one when it is not named. Every agent named must be one a PROCESS stage of the shape runs, and every revision must exist — checked at each step, so a walk is refused at its first.
Run the caller's draft of the stage's agent instead of a published revision — the edit being made, tested on the stage before it is even a revision. Held to every check a revision would be. Refused at an ASK stage, and together with CreateLedgerStageTestRequest.AgentRevision or CreateLedgerStageTestRequest.AgentRevisions.
Response Body
curl -X POST "https://api.fruxon.com/v1/tenants/string/triggers/497f6eca-6276-4993-bfeb-53cbbbba6f08/stageTests" \ -H "Content-Type: application/json" \ -d '{}'{
"id": "00000000-0000-0000-0000-000000000000",
"walkId": "00000000-0000-0000-0000-000000000000",
"step": 0,
"previousTestId": "00000000-0000-0000-0000-000000000000",
"stageIndex": 0,
"stageId": "string",
"stageName": "string",
"status": "QUEUED",
"seedKind": "ITEM",
"seedItemId": "00000000-0000-0000-0000-000000000000",
"replayOfExecutionRecordId": "00000000-0000-0000-0000-000000000000",
"itemKey": "string",
"fromDraft": false,
"input": null,
"userQuery": "string",
"reply": "string",
"agentId": "string",
"agentRevision": 0,
"agentRevisionDeployed": false,
"agentDraftBaseRevision": 0,
"routingDefault": "NONE",
"routingOverrides": {},
"executionRecordId": "00000000-0000-0000-0000-000000000000",
"output": "string",
"verdict": {
"kind": "ADVANCE",
"toStatus": "READY",
"cause": "string",
"detail": "string",
"failedCheck": "RUN",
"attempt": 0,
"maxAttempts": 0,
"nextStageId": "string",
"routed": false,
"produced": null,
"dataAfter": null,
"dedupeKey": "string",
"dedupeKeyIncomplete": false,
"duplicateOf": "string",
"nextStageMissing": [
"string"
],
"sink": {
"stageId": "string",
"kinds": [
"UNSPECIFIED"
],
"target": "string"
}
},
"simulatedCalls": [
{
"integrationId": "string",
"toolKey": "string",
"operation": "CREATE",
"simulatedEntityId": "00000000-0000-0000-0000-000000000000",
"createdAt": 0
}
],
"writesKept": false,
"cost": 0,
"durationMs": 0,
"warnings": [
"string"
],
"createdAt": 0,
"completedAt": 0,
"expiresAt": 0
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}{
"type": "string",
"title": "string",
"status": 0,
"detail": "string",
"instance": "string",
"property1": null,
"property2": null
}Stops a stage test whose run has not ended.
Cancels the run and keeps the test, so what was started stays readable. Returns 409 when the test has already ended.
Lists recent source events for an unsaved trigger draft.
Returns the last N items the source mailbox or workspace would have matched, annotated with whether the draft filter would have fired on each. This backs the create-flow panel before a trigger row exists. It is read-only — no upstream writes, no trigger persistence, no agent enqueue. The capability is per-`(integration, eventType)`; pairs without a registered `IRecentEventSource` return 400.`.