[JacobStephens2]

Summary

- Count review observations, including notes accepted as-is or declined.
+ List and count only actionable findings requiring a change or human follow-up.
+ Record 0 for an axis with no actionable findings.

Define the same contract in the Session prompt and embedded Factory skill, regenerate the published prompts, and align the manual and hand-written process page. Findings later fixed still count. The JSON schema and retained historical metrics stay unchanged.

Closes #2051

Evidence

Execution evidence

Confirmed cause: the shared report prompt omitted an actionable-only definition, while the Factory skill allowed an unqualified finding count and explicitly included declined findings.

• Before: cargo test --bin thirdshift prompt::tests::review_reports_count_only_actionable_findings failed: missing "number of actionable findings".
  After: the same command passed (1 test).
• Before: cargo test --bin thirdshift skills::tests::code_review_reports_count_only_actionable_findings failed: missing "Report only actionable findings that warrant a change or human follow-up" in Factory skill.
  After: the same command passed (1 test).
• Before: node --test site/tests/review-counts.test.cjs failed: docs/guide.md: stale count contract.
  After: the same command passed (1 test). The Spec reviewer's exact Python assertion also failed for all three published documentation paths before the fix and passed afterward.
• UPDATE_PROMPTS=1 cargo test prompts_page and cargo test prompts_page: passed (14 prompt-page tests each).
• npm run docs --prefix site and npm run docs:check --prefix site: passed.
• scripts/affected-tests refs/remotes/origin/main: site checks passed (50.70s); its Rust run passed the unit tests and earlier integration targets, then caught the stale exact-prompt expectation in fresh_run (528.31s). Updated that expected text; cargo test --test fresh_run passed (26 tests, 2 ignored). Continuing the 77 integration targets that the first run did not reach; completed targets use identical production sources. The continuation passed all 77 remaining targets (1039.94s, warm build cache). Together with the corrected fresh-run file, all integration targets and the unit suite are covered successfully; no equivalent full-suite repeat was needed.

The first scripts/affected-tests main attempt selected changes from 49 commits already present on the fetched base, because local main was stale. Its extra Console suite failed at working-targets.test.cjs:83 (1440px remote review-explicit: visible PR target, false !== true); a direct node --test console/tests/working-targets.test.cjs rerun failed identically. No Console source or test is changed by this PR. The final affected selection uses the fetched PR base, refs/remotes/origin/main.

Test seams: emitted Session prompts, the Factory skill linked into a worktree, and the published review-count documentation. These deterministic checks guard the instruction contract; they do not measure a model's compliance rate.

Visual evidence

Prompts count contract, 1440×900

Prompts count contract, 390×844

Manual count contract, 1440×900

Manual count contract, 390×844

How it works count contract, 1440×900

How it works count contract, 390×844

Review coverage

Standards: 0 findings. Spec: 1 finding, fixed in 98127183; none remain. Both reviewers read the 20 files present at 98127183, including generated files. Their one follow-up was already complete when validation exposed the stale exact-prompt expectation. The later tests/fresh_run.rs expectation update is unread by both review axes and is disclosed here; it passed its full integration-test file. No generated-output exclusions were used.

Unaddressed findings

Standards

None.

Spec

None. The published-documentation finding was reproduced and fixed, with its regression retained.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Revert the instruction and documentation changes to restore the previous behavior.

Blast Radius: reviews

Future review Session prompts and documentation only; no Rust count schema change or rewrite of historical summaries.

Built with codex codex-cli 0.167.2 · gpt-6.1-sol · xhigh