SpecSmith

SpecSmith is an agentic QA system that helps teams find what they forgot to test before bugs reach production. Developers often build from PRDs, GitHub issues, and OpenAPI docs, but test planning is still manual, inconsistent, and easy to miss. SpecSmith turns those raw specs into a structured Forge Report with risk-ranked findings, a linked test matrix, executable test drafts, a coverage score, and a gap report. The product is built around a QA workflow with specialized agents: Spec Parser, Risk Mapper, Test Planner, Test Writer, and QA Reviewer. The public demo uses a fast OpenAI-compatible API mode optimized for reliability: it makes one real model call, reads the actual input spec, and generates a context-aware Forge Report. The full 5-agent pipeline is also implemented and validated separately, including Qwen 72B validation evidence documented in the repository. SpecSmith is not just a generic test generator. It connects requirements, risks, tests, and coverage gaps so engineering teams can understand what matters before they ship. The AMD Developer Cloud path is documented for a future vLLM/Qwen deployment on MI300X once credits are approved.