Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Render the patch-edit run's accumulated changes as a reviewable diff, surface it through a GenUI choice surface, and persist the user's accept / reject decision into the artifact manifest.
.claude/skills/nexu-io-diff-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 2% | 0% |
Spec §20.3 / §21.3.2: a code-migration / tune-collab handoff is worthless without a human-reviewable diff. This atom is the "present the patch, capture the decision" stage.
since the previous diff-review iteration).
plan/steps.json for context per file.textproject-cwd/ └── review/ ├── diff.patch # unified diff (git apply-shaped) ├── summary.md # human-friendly per-file walkthrough ├── decision.json # { decision: 'accept' | 'reject' | 'partial', accepted_files: [...], rejected_files: [...], reviewer: 'user' | 'agent' } └── meta.json # { generatedAt, atomDigest, planRevision }
The atom raises a choice GenUI surface with the three top-level decisions (accept / reject / partial). On partial the user flips per-file decisions through follow-up surfaces.
The atom completes when decision.json has a non-empty decision. Acceptance writes handoffKind: 'patch' (or 'deployable-app' when a successful build-test is on file) into the eventual artifact manifest; rejection rolls back the patch via git restore or the equivalent in the code-import-bound repo path.
decision.json that lacks accepted_files /rejected_files on partial decisions.
plan/steps.json's union offiles[] — the patch boundary is a contract.
Implemented by the daemon runner in apps/daemon/src/plugins/atoms/diff-review.ts, with GenUI decisions persisted by apps/daemon/src/plugins/atoms/diff-review-genui-bridge.ts.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 8,566 | 7,169 | -16% | 1 | 1 | 0% | 1,340 | 1,453 | +8% | 0 | 0 | — |
case-01 | fail→fail | 6,449 | 10,811 | +68% | 1 | 1 | 0% | 280 | 687 | +145% | 0 | 0 | — |
case-02 | fail→fail | 6,576 | 8,442 | +28% | 1 | 1 | 0% | 269 | 1,026 | +281% | 0 | 0 | — |
case-03 | fail→fail | 16,836 | 8,913 | -47% | 1 | 1 | 0% | 473 | 1,155 | +144% | 0 | 0 | — |
case-05 | fail→fail | 8,710 | 1,874 | -78% | 1 | 1 | 0% | 1,505 | 840 | -44% | 0 | 0 | — |
case-06 | fail→pass | 11,207 | 1,931 | -83% | 1 | 1 | 0% | 1,521 | 810 | -47% | 0 | 0 | — |
case-07 | fail→pass | 8,385 | 4,516 | -46% | 1 | 1 | 0% | 1,374 | 1,135 | -17% | 0 | 0 | — |
case-08 | pass→pass | 16,135 | 4,876 | -70% | 1 | 1 | 0% | 2,146 | 1,153 | -46% | 0 | 0 | — |
case-09 | pass→pass | 11,498 | 4,188 | -64% | 1 | 1 | 0% | 1,966 | 1,162 | -41% | 0 | 0 | — |
case-10 | fail→pass | 14,149 | 4,868 | -66% | 1 | 1 | 0% | 2,869 | 1,321 | -54% | 0 | 0 | — |
case-11 | fail→pass | 23,464 | 13,231 | -44% | 1 | 1 | 0% | 3,068 | 2,391 | -22% | 0 | 0 | — |
case-12 | fail→pass | 14,927 | 10,580 | -29% | 1 | 1 | 0% | 2,290 | 2,334 | +2% | 0 | 0 | — |
case-13 | pass→pass | 11,422 | 13,492 | +18% | 1 | 1 | 0% | 1,793 | 2,396 | +34% | 0 | 0 | — |
case-14 | fail→fail | 15,095 | 15,706 | +4% | 1 | 1 | 0% | 2,163 | 2,806 | +30% | 0 | 0 | — |
case-15 | fail→pass | 13,280 | 3,488 | -74% | 1 | 1 | 0% | 1,977 | 1,071 | -46% | 0 | 0 | — |
case-16 | pass→pass | 6,513 | 3,788 | -42% | 1 | 1 | 0% | 864 | 1,071 | +24% | 0 | 0 | — |
case-17 | fail→pass | 13,033 | 4,065 | -69% | 1 | 1 | 0% | 1,697 | 984 | -42% | 0 | 0 | — |
case-18 | pass→pass | 14,535 | 3,146 | -78% | 1 | 1 | 0% | 2,065 | 888 | -57% | 0 | 0 | — |
case-19 | fail→pass | 44,152 | 1,449 | -97% | 1 | 1 | 0% | 2,531 | 719 | -72% | 0 | 0 | — |
case-20 | fail→fail | 5,091 | 3,281 | -36% | 1 | 1 | 0% | 211 | 906 | +329% | 0 | 0 | — |
case-21 | fail→fail | 2,216 | 17,109 | +672% | 1 | 1 | 0% | 331 | 2,768 | +736% | 0 | 0 | — |
case-22 | fail→fail | 11,568 | 4,498 | -61% | 1 | 1 | 0% | 149 | 680 | +356% | 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. 22 cases were attempted, and 17 counted toward the lift figure. The other 5 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 +36 percentage points is the difference between those two pass rates over the 17 comparable cases.
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.
Other measured skills in the registry, with their headline benchmark lift.