generate_feature

Dispatch AI test generation for a feature. Returns immediately with a requestId; the pipeline runs in a cloud sandbox for several minutes.

warning

Generation consumes account credits. One generation per feature at a time — a second dispatch while one is in flight returns 409 conflict.

Request

POST /api/v1/ops/projects/{project}/features/{feature}/generate

Parameter	In	Type	Required	Description
project	path	string	yes	The project's public id.
feature	path	string	yes	Feature id or name.
siteUrl	body	string	no	Test site URL override (defaults to the project's configured URL).

cURL

curl -X POST -H "Authorization: Bearer $TESTVIBE_API_KEY" \
  -H "Content-Type: application/json" -d '{}' \
  "$TESTVIBE_SERVER/api/v1/ops/projects/$PROJECT/features/12/generate"

CLI

testvibe generate "Login Functionality" --watch --timeout 3600

MCP

Tool generate_feature — { "feature": "Login Functionality" }. Follow with wait_for_generation.

Response

202 Accepted

{ "requestId": "a1b2c3d4-…", "featureId": 12 }

The feature's status flips to generating; track progress via get_generation_status and branch on isTerminal, terminalReason, or awaitingInput.

info

--watch follows it to the end and exits 0 only when the generation actually succeeded — 1 if it failed or was cancelled, 2 if it parked to ask you a question, 3 if --timeout (default 5400 s) elapsed first. See the CLI wait contract.

Errors

HTTP	When
404 not_found	No such feature in this project.
409 conflict	Already generating, or the dispatch was rejected.

Additional dispatch options

Body field	Type	Behavior
model	string	Exact id from GET …/projects/{project}/models / list_models / testvibe models; omit for the project default, or use auto for model selection. Applies only to this generation.
gherkinChanges	string	ask (default), accept, deny, or stop. Unknown values fall back to ask.
notify	boolean	Default false; opt into normal completion notifications. CLI: --notify.
preflight	boolean	REST/MCP: return the target, mutation risk, references, and confirmation information without starting generation or choosing a model. Returns 200.
confirm	string	REST/MCP: echo the preflight token when the deployment's transactional confirmation gate requires it. A source edit invalidates the token.

Generation drives the target application live. Its steps can change target data during generation, before a separately dispatched test run. A required confirmation returns 409 confirmation_required; unknown models return 400 invalid_model. Do not hard-code the model catalog: inspect ids and configured on the current installation. Model selection responses can add model, modelSource, and modelReason.

Answer a paused generation

Poll generation status. If awaitingInput is true, send exactly one answer shape to:

POST /api/v1/ops/projects/{project}/features/{feature}/generation/answer

The body is {requestId?, option}, {requestId?, message}, or {requestId?, fields:[{name,value}]}. Prefer the question's pendingInput.requestId when present. Do not substitute an unrelated workflow request id. The response is 202 with ok, requestId, featureId, applied, and decision; resumed work can have a newRequestId.

testvibe generate answer 12 --option "<exact option from pendingInput>" --watch

MCP uses answer_generation with feature, optional requestId, and option, message, or fieldsJson (a JSON-encoded array). An already answered/cancelled question returns 409. Inspect a resume_unavailable or resume_failed response before deciding whether fresh generation is required.