Deterministic no-LLM runner for tests and CI (spec §4).
Calls no model. It echoes the rendered tool (name, resolved description, and
the full wire map) plus the eval prompt back as the transcript, so the harness
pipeline — and "the rendered description varies per permutation" assertions —
can run without network access. Pairs with Noizu.MCP.Eval.Judge.Stub.
Real LLM runners are app-layer follow-ups; select one via the :noizu_mcp
:eval_runner application env.