[JacobStephens2]

Summary

affected-tests <base>
  site       -> browser + Rust site checks
  release    -> release + generator tests
  mixed/other -> full Rust suite (nextest, cargo test fallback)
                 + selected browser/generator checks

Document repository test selection and binary-only selectors. The Factory implement skill follows repository guidance and retains the full-suite fallback elsewhere.

Closes #724

Evidence

• Before: python3 -m unittest discover -s scripts -p test_affected_tests.py failed for missing site/release selection and for mixed changes dropping browser checks.
  After: all 12 selector tests pass, covering site, release, Rust, support/fakes, mixed, unknown, dirty files, rename, fallback and failures. All 14 tests pass with python3 -m unittest discover -s scripts -p 'test_*.py'.
• cargo check --all-targets passes. UPDATE_PROMPTS=1 cargo test prompts_page regenerates only the expected skill paragraph; cargo test prompts_page passes all 11 tests afterward. npm test --prefix site passes all 36 browser tests on the regenerated page.
• /usr/bin/time -p cargo nextest run (nextest 0.9.146, default profile, Linux, warm build): 2,223 passed, 2 skipped, across 52 binaries; 1,295.90s wall time (21m35.90s), including 0.32s reported build time. Nextest's test-only summary was 1,294.215s. This is a measurement of this host, not a demonstrated speedup over cargo test.
• CI run 39687711646 passed on final head f598872, including the Linux Rust suite, tooling checks, site/Console browser tests and macOS gate. The already-running local measurement completed without starting another full run.
• Local selection, nextest setup/fallback, binary-only selectors and broader CI responsibilities are documented in docs/testing.md. CI workflow and Delivery behavior are unchanged.

Test seams

• scripts/affected-tests <base> as a subprocess in a real scratch Git repository, observing exit status, printed selection, and external test-runner invocations.
• Existing generated-prompts consistency and site browser tests.

Review coverage

Independent Standards and Spec reviews against main found no issues. Each axis read all six changed files, including the generated page's complete changed hunk. Unread changed files: none. Generated-output exclusions: none. Reports with each axis's final supporting-files list are retained in .thirdshift-review-mzw1W6/standards.md and .thirdshift-review-mzw1W6/spec.md in the worktree.

Unaddressed findings

Standards

None.

Spec

None.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Blast Radius: validation

Conservative local selection and Factory testing guidance; CI and Delivery policy remain unchanged.

Built with codex · gpt-6.1-sol · high