Behaviour implemented by host applications to deliver Squid Mesh work.

Squid Mesh owns workflow state, retry decisions, and progression. The host executor owns how runnable work is queued and later delivered back to the runtime entrypoints in SquidMesh.Runtime.Runner.

A typical executor converts the run intent into SquidMesh.Executor.Payload data, stores that payload in the host app's job backend, and returns metadata about the queued job.

def enqueue_step(_config, run, step, opts) do
  job = %{
    payload: SquidMesh.Executor.Payload.step(run, step),
    queue: queue(),
    schedule_in: opts[:schedule_in]
  }

  with {:ok, queued_job} <- MyApp.JobQueue.enqueue(job) do
    {:ok, %{job_id: queued_job.id, queue: queued_job.queue}}
  end
end

The queued job should call SquidMesh.Runtime.Runner.perform/1 with the stored payload. The job backend, queue name, retry settings, and scheduler stay in the host application.