Create a Worker task using an explicit workflow version¶
POST /workspaces/{id}/tasks is a Worker compatibility endpoint calling the shared workflow execution service. Exactly one nonempty X-Workspace-ID must match the path workspace. It creates neither automation rules nor schedules.
Request body¶
{
"task": {
"name": "sample-run",
"workflow_version_id": "version-example",
"data": "{\"text\":\"sample\"}",
"vars": "{}"
}
}
Fields |
Type |
Required |
Description |
|---|---|---|---|
|
object |
Yes |
Object described below |
|
string |
Yes |
The handler requires nonempty input; no global uniqueness or surrounding-whitespace rules are implied |
|
string |
Yes |
Explicit published version ID in the current workspace, not a workflow application ID |
|
string |
No |
Encoded JSON object string. Omitted/blank means an empty object. Arrays, null, and invalid JSON are rejected |
|
string |
No |
Also an encoded JSON object string, not a direct JSON object |
The body permits one complete JSON value and rejects unknown fields. cron_expression, transient, trace, or other extra fields in the body/task trigger the strict DTO unknown-field rejection. Do not describe unknown fields as ignored.
Execution conditions and side effects¶
The version must resolve to exactly one workflow application with a deployment definition and latest-version reference. The target version must belong to that definition, be published, and have nonempty workflow content. Suspended/disabled applications reject execution. Dynamic-service versions still validate the input schema.
The permission boundary resolves application/resource facts, checks the principal, authorizes, and injects the effective role. Selecting a task name and version uses one_shot delivery. Even a Cron application creates an independent single-run child execution rather than creating/replacing a schedule. Full resource discovery and runtime policy still require review; one service test does not establish authorization of every Catalog dependency.
The shared execution chain prepares Mowl task/case, persists execution and lineage in a transaction, and reads them back. It wakes execution only after commit. A wake-up failure may occur after records have committed. Query existing runs before resubmitting. There is no request idempotency key or guaranteed retry deduplication.
Successful response¶
{
"task": {
"id": "worker-task-example",
"name": "sample-run",
"workflow_version_id": "version-example",
"workflow_execution_id": "execution-example"
}
}
HTTP 200 returns task directly, without an outer code/msg/data envelope:
Field |
Type |
Subsequent use |
|---|---|---|
|
string |
Mowl/Worker task ID, not an agent authoring task ID |
|
string |
Requested task name |
|
string |
Requested explicit version ID |
|
string |
Product workflow execution ID for execution details and results |
There is no response state field. Worker protobuf zero values or client defaults do not prove running, completion, or successful output. Common parameter and undeployed errors return 400/ErrParamInvalid; missing workspace returns 400/ErrWorkspaceRequired. Other authorization, resource, execution-identity, dependency, storage, and wake-up errors follow shared workflow mappings.