manage_workflows

CRUD and execution for visual workflows. Actions: list, get, create, update, delete, run, approve, reject, publish. Terminology: In Scenario, when a user says "apps" (e.g. "list my apps"), they usually mean workflows in "ready" status — use action="list" with status="ready". Same idea for "my ready apps" or "deployable workflows". Lifecycle: create / update persist editor_info + inputs as a DRAFT; they do NOT make the workflow runnable. To go from a draft (editor state) to a ready workflow that action="run" can invoke, call action="publish" — it compiles editor_info into a flow and flips status to "ready". The split mirrors the webapp's Save (persist) vs. Convert-to-App (publish) distinction. Importing a webapp JSON export is therefore a TWO STEP call: 1) create with name + editor_info + inputs_definition, 2) publish with the returned workflow_id. Args: - action: required operation (see above) - workflow_id: required for get, update, delete, run, approve, reject, publish - limit, page_token: pagination for list (default limit=20, max 200). Response carries nextPaginationToken when more pages are available. - status: optional for list only. Values "draft" and "ready" are server-filtered (the API supports them natively, faster). Any other value (e.g. "running", "archived") falls back to a client-side filter applied to the returned page only — _workflowListStatusFilter metadata flags this case. - name, description: used for create or update - flow, inputs_definition, editor_info: typed workflow definition fields, accepted on both create and update. On create the workflow is provisioned first then patched with these fields, so a single create call can clone a workflow. editor_info is validated against the webapp's import schema (must include non-empty nodes / edges arrays and recognized node types); create / update fail with the validator's message rather than persisting a broken state. - inputs, dry_run: used for run. dryRun lands in the URL query string (the SDK splits it from the body), inputs becomes the request body. For array-typed inputs (file_array, string_array, model_array), you may pass a bare scalar value and it will be wrapped into a single-element array before the call — the API's strict validator drops scalar-for-array mismatches silently. - workflow_job_id, node_id: required for approve and reject - team_id, project_id: required for OAuth callers Returns: workflow metadata for CRUD calls, workflow run results for run (typed WorkflowRunResponse with .job for tracking), and approval results for approve/reject. action=list returns a typed workflows array plus optional nextPaginationToken; when a non-API status filter is applied, the response also includes _workflowListStatusFilter. When create / update persist editor_info without a precompiled flow, the response includes _publishHint reminding the caller to chain action="publish". Examples: - "List workflows" -> action="list" - "List my apps" / "ready workflows" -> action="list", status="ready" - "Run this workflow" -> action="run", workflow_id="workflow_xxx", inputs={...} - "Estimate cost without running" -> action="run", workflow_id="workflow_xxx", inputs={...}, dry_run=true - "Import this workflow JSON and make it runnable" -> action="create" with name + editor_info + inputs_definition, then action="publish" with the returned workflow_id - "Approve this workflow node" -> action="approve", workflow_id="workflow_xxx", workflow_job_id="job_xxx", node_id="node_xxx" Don't use when: You only need a single model invocation. Prefer run_model instead.

{
  "action": "run",
  "workflow_id": "wf_hero_pipeline",
  "inputs": {
    "prompt": "epic fantasy hero portrait",
    "style_model_id": "model_custom_abc"
  },
  "team_id": "team_abc123",
  "project_id": "proj_xyz789"
}

{
  "workflow": {
    "id": "wf_hero_pipeline",
    "status": "running",
    "jobId": "job_wf_001"
  }
}