Data Machine exposes datamachine/execute-agent-workflow as the supported bridge for simple Agents API workflow specs.

The bridge accepts a WP_Agent_Workflow_Spec array, passes execution to WP_Agent_Workflow_Runner, and records the run as a Data Machine direct job. It does not convert the spec into a Data Machine pipeline or implement a second workflow runner.

Supported Scope

Supported workflow shape:

• top-level ability steps;
• top-level agent steps, via the Agents API agents/chat ability;
• empty triggers or on_demand triggers;
• runner inputs, metadata, run IDs, and continue_on_error.

Unsupported workflow features fail before execution with a typed error. Data Machine does not bridge foreach, cron triggers, action triggers, branching, parallel execution, nested workflows, or consumer-defined step types through this ability.

Job Mapping

Each run creates one Data Machine job with:

• pipeline_id: direct;
• flow_id: direct;
• source: agents_api_workflow;
• label: caller-provided label or Agents API Workflow: <workflow_id>.

The job engine_data stores:

• agents_api_workflow.workflow_id and agents_api_workflow.run_id;
• the source spec, runner inputs, and runner metadata;
• workflow_run_result, mirroring WP_Agent_Workflow_Run_Result::to_array();
• step_outcomes, output, and error for job reporting;
• artifacts and logs, copied from metadata.artifacts and metadata.logs when the caller supplies them because WP_Agent_Workflow_Run_Result has no separate first-class artifact/log fields;
• provenance identifying Agents API as the source, WP_Agent_Workflow_Runner as the executor, and Data Machine jobs as the recording surface.

Runner status maps to Data Machine job status as follows:

• running remains processing;
• succeeded becomes completed;
• failed becomes failed - <error_code>;
• skipped becomes failed - agents_api_workflow_skipped.