amiya-special/dynaflow is a Python project that deals with the way a model either answers directly or generates and runs an agent task flow. The repository description introduces it as a project for learning free-form multi-agent orchestration. Rather than following a fixed-order workflow, it looks at how a model composes a collaborative flow among several agents. FACTS lists no separate release version.
According to the README summary, the comparison targets are three execution modes: direct, dynamic, and forced. direct answers without tools, dynamic chooses between answering directly and running a flow, and forced requires a flow execution in which the handoff of the task to a sub-agent succeeds. The project stores the answer, the call log, the comparison material, and the flow record as execution results. So it does not stop at merely listing agent calls; it records and compares the difference between direct responses and flow execution. FACTS also explains that it provides a real API demo and a workflow handoff verification.
However, no separate test suite could be confirmed in the repository tree. The demo and the handoff verification are grounds for the implementation approach, but that differs from evidence that the answer quality of the several modes was compared systematically. This project can therefore be viewed as an implementation for composing and observing multi-agent flows, but the material for judging that those flows are better than direct responses is limited.