The project runs one test-first pipeline with three selectable review-rigor variants, brokered by LiteLLM (so the reviewer model can be any provider — hosted or fully local):
basesecond-opinion(also called dual review)panel
For the details — how each works, how to run it, and the proof the gates catch defects — go straight to the dedicated pages:
- Running the Pipeline — set up and drive each variant.
- Pipeline Verification — a worked proof the gates catch defects.
- Coding Pipeline — the phases, roles, and per-variant flow diagrams.
