Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Independent cross-model review of the current review scope (branch diff plus staged, unstaged, untracked changes) from the opposite vendor's model. Catches spec drift, missing edge cases, and fake tests self-review cannot see. Use before merging, after a tricky change, or a debug second opinion.
.claude/skills/ancienttwo-repo-harness-cross-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 87% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -58% | 0% |
Canonical rule owner for independent outside review. Scope capture, provider invocation, timeout, structured-output validation, and error classification live in code (src/core/review, src/effects/review, src/cli/commands/cross-review.ts); this package owns only when to invoke, how to interpret findings, and the boundaries below.
references/codex-mode.md.codex@openai-codex plugin app-server: references/codex-plugin-mode.md.references/claude-mode.md.[P1] finding -> FAIL (do not merge until addressed). Only [P2] or none -> PASS.SKIPPED -- advisory, non-blocking (exit 0). Do not re-run it or narrow the diff to retry. degraded_scope and stale_scope block (exit 1). Never a synthesized pass.merge-gate receipt; that judge is separate.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 13,887 | 4,079 | -71% | 1 | 1 | 0% | 2,143 | 1,181 | -45% | 0 | 0 | — |
case-04 | fail→pass | 12,673 | 2,400 | -81% | 1 | 1 | 0% | 2,183 | 772 | -65% | 0 | 0 | — |
case-01 | fail→fail | 5,454 | 5,405 | -1% | 1 | 1 | 0% | 895 | 770 | -14% | 0 | 0 | — |
case-02 | fail→fail | 7,281 | 8,839 | +21% | 1 | 1 | 0% | 1,108 | 678 | -39% | 0 | 0 | — |
case-03 | fail→fail | 6,456 | 8,367 | +30% | 1 | 1 | 0% | 946 | 1,031 | +9% | 0 | 0 | — |
case-06 | pass→pass | 4,726 | 2,438 | -48% | 1 | 1 | 0% | 707 | 856 | +21% | 0 | 0 | — |
case-07 | fail→pass | 3,249 | 20,294 | +525% | 1 | 1 | 0% | 541 | 1,012 | +87% | 0 | 0 | — |
case-08 | pass→pass | 7,753 | 2,872 | -63% | 1 | 1 | 0% | 1,252 | 850 | -32% | 0 | 0 | — |
case-09 | pass→pass | 8,341 | 2,010 | -76% | 1 | 1 | 0% | 1,397 | 729 | -48% | 0 | 0 | — |
case-10 | fail→pass | 10,879 | 4,844 | -55% | 1 | 1 | 0% | 1,749 | 1,183 | -32% | 0 | 0 | — |
case-11 | fail→pass | 42,306 | 2,483 | -94% | 1 | 1 | 0% | 2,050 | 866 | -58% | 0 | 0 | — |
case-12 | fail→pass | 11,161 | 3,308 | -70% | 1 | 1 | 0% | 1,762 | 983 | -44% | 0 | 0 | — |
case-13 | fail→pass | 9,831 | 3,300 | -66% | 1 | 1 | 0% | 1,567 | 984 | -37% | 0 | 0 | — |
case-14 | fail→pass | 13,735 | 4,096 | -70% | 1 | 1 | 0% | 2,251 | 1,041 | -54% | 0 | 0 | — |
case-15 | fail→pass | 22,598 | 2,464 | -89% | 1 | 1 | 0% | 1,619 | 726 | -55% | 0 | 0 | — |
case-16 | pass→pass | 10,952 | 4,520 | -59% | 1 | 1 | 0% | 1,753 | 1,152 | -34% | 0 | 0 | — |
case-17 | pass→pass | 10,478 | 4,257 | -59% | 1 | 1 | 0% | 1,653 | 972 | -41% | 0 | 0 | — |
case-18 | pass→pass | 12,596 | 11,229 | -11% | 1 | 1 | 0% | 1,994 | 1,250 | -37% | 0 | 0 | — |
case-19 | fail→pass | 7,284 | 2,475 | -66% | 1 | 1 | 0% | 1,170 | 815 | -30% | 0 | 0 | — |
case-20 | pass→pass | 19,106 | 16,194 | -15% | 1 | 1 | 0% | 4,199 | 4,083 | -3% | 0 | 0 | — |
case-21 | fail→fail | 30,059 | 3,256 | -89% | 1 | 1 | 0% | 1,171 | 985 | -16% | 0 | 0 | — |
case-22 | pass→pass | 5,745 | 7,539 | +31% | 1 | 1 | 0% | 1,287 | 2,050 | +59% | 0 | 0 | — |
case-23 | pass→pass | 10,977 | 12,375 | +13% | 1 | 1 | 0% | 2,478 | 3,309 | +34% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 19 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +43 percentage points is the difference between those two pass rates over the 19 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v3, not the version currently listed.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/21/2026 | +36% |
Other measured skills in the registry, with their headline benchmark lift.