ARTIFACTS → RETRIEVAL → AGENTS → REVIEW → EVIDENCE
Turn product evidence into reviewable test intelligence.
Bring requirements, APIs, tests, screenshots and logs together. Each bounded agent leaves a traceable output for human review.
Evidence intake
Attach metadata for the evidence you want considered. Files stay local in this reference; only filenames, types and sizes are stored. Repository execution is never enabled by an upload.
Requirements agent
Only bundled expense policies and nine catalog tests are executable in this reference implementation.
Agent review & approval
Generate or select a plan to review the policy sources and test selection.
Agent architecture
Generate a plan to view requirement analysis, traceability, Gherkin, generated code and review checks.
Sandboxed execution & failure analysis
The execution agent runs only approved catalog tests against isolated synthetic fixtures. On failure, it correlates the selected requirement and result into a draft defect; submission always needs human approval.