This walks through triggering a single agent via a minimal workflow, assuming you’ve completed Quickstart.
1

Confirm the agent is registered

If make seed ran successfully, this returns the agent’s manifest — agent_id, entrypoint, category, and (if assigned) its evaluator.
2

Trigger the workflow

Returns a run_id immediately — execution is asynchronous, dispatched to a Celery worker.
3

Watch it move

Poll (or use the dashboard’s live WebSocket view) to watch the run move PENDING → RUNNING → COMPLETED, with each step’s own state alongside it.
4

Inspect what happened

The run response includes each step’s output, cost (cost_usd), and — if the agent has an evaluator assigned — its evaluation score. No evaluator assigned means null score and "evaluated": false, not a default pass. See AI Agents → Evaluation.

​
What actually happened

  1. The API validated your request and inserted a workflow-run row
  2. The orchestrator resolved the DAG (one node here) and dispatched a Celery task to the orchestration/agent_high queue
  3. LifecycleManager.run() acquired the ADR-006 idempotency lock, resolved the agent’s manifest entrypoint, and executed it
  4. The agent’s real body ran — a tool call to the Ahrefs connector, then an LLM call for intent classification, then deterministic opportunity scoring
  5. The output was written to session memory, then long-term memory; a lifecycle event was emitted; the lock was released

Content pipeline walkthrough