Read and apply the canonical test-value-evidence skill first, then these repository-specific requirements. Keep the canonical vendored file unmodified; its source revision and blob are recorded in eng/vendored-files.json.
- Read repository instructions. Use the
repository-pinned toolchain through
build.cmdorbuild.sh, capture binary logs for MSBuild commands, and use the owning project's assertion conventions andBannedSymbols.txt. MSTest framework tests useTestContainer; MTP and analyzer tests use MSTest; adapter tests use their established assertion APIs. - Apply parallel-safety preflight before test/lifecycle changes and against the final diff. Counterfactual experiments must use isolated outputs and preserve the actual test scheduling.
- For acceptance, package, launcher, protocol, or cross-component changes, apply acceptance validation. Repack each product variant before testing, prove the effective package/cache and loaded binary paths, and retain real shipping-layout evidence. Assert/API overload changes also need the repository's packed source-compatibility coverage.
- Reuse the installed test-gap-analysis for bounded static reasoning and the existing mutation workflow for available executed Stryker evidence. Do not upgrade unrelated curated plugins to run this skill.
- Keep publication under the code-review gate.
The automatic and
/review-testsreviewers are read-only: report static predictions or unavailable proof unless exact matching execution records can be inspected. Do not execute PR code or infer red-green proof from green CI. Put material evidence gaps in the summary, not automatic inline findings.
The existing review workflow reports the canonical evidence table for each eligible PR head, including unchanged tests protecting source-only changes. Individual commit validation is separate; neither the skill nor its vendored presence enforces a Git hook or proves live workflow publication.