Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Default reference pipeline for the code-migration taskKind — code-import → design-extract → token-map → rewrite-plan → patch-edit ↔ build-test devloop → diff-review → handoff.
.claude/skills/nexu-io-od-code-migration/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -53% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -67% | 0% |
Spec §1 / §10.1 / §20.3 / §21.3.2 / §23.3.3: the canonical code-migration flow. The pipeline cannot ship an accept decision without a passing build — the patch-edit ↔ build-test devloop is the convergence engine.
jsonc{ "stages": [ { "id": "import", "atoms": ["code-import"] }, { "id": "tokens", "atoms": ["design-extract", "token-map"] }, { "id": "plan", "atoms": ["rewrite-plan"] }, { "id": "verify", "atoms": ["patch-edit", "build-test"], "repeat": true, "until": "(build.passing && tests.passing) || iterations>=8" }, { "id": "review", "atoms": ["diff-review"] }, { "id": "handoff", "atoms": ["handoff"] } ] }
The scenario uses build.passing && tests.passing as the convergence signal — promoted into the spec §22.4 vocabulary by the build-test atom.
A plugin built on this scenario typically declares od.inputs:
repoPath (string, required): pulled from od project import.targetStack (form): collected by rewrite-plan.testCommand / buildCommand (string, optional): overridesthe inferred package.json scripts in the build-test atom.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,615 | 27,990 | +79% | 1 | 1 | 0% | 3,282 | 2,084 | -37% | 0 | 0 | — |
case-02 | fail→pass | 43,546 | 7,779 | -82% | 1 | 1 | 0% | 3,038 | 1,615 | -47% | 0 | 0 | — |
case-03 | fail→pass | 21,127 | 5,657 | -73% | 1 | 1 | 0% | 4,268 | 1,482 | -65% | 0 | 0 | — |
case-04 | pass→pass | 10,250 | 7,753 | -24% | 1 | 1 | 0% | 1,819 | 1,894 | +4% | 0 | 0 | — |
case-05 | pass→pass | 12,702 | 11,087 | -13% | 1 | 1 | 0% | 2,423 | 1,978 | -18% | 0 | 0 | — |
case-06 | pass→pass | 14,295 | 8,063 | -44% | 1 | 1 | 0% | 2,794 | 1,993 | -29% | 0 | 0 | — |
case-07 | fail→pass | 17,108 | 3,498 | -80% | 1 | 1 | 0% | 2,341 | 1,108 | -53% | 0 | 0 | — |
case-08 | fail→pass | 31,996 | 2,898 | -91% | 1 | 1 | 0% | 2,279 | 742 | -67% | 0 | 0 | — |
case-09 | fail→pass | 11,330 | 5,528 | -51% | 1 | 1 | 0% | 1,409 | 762 | -46% | 0 | 0 | — |
case-10 | fail→pass | 17,039 | 3,476 | -80% | 1 | 1 | 0% | 2,261 | 847 | -63% | 0 | 0 | — |
case-11 | fail→pass | 10,849 | 1,894 | -83% | 1 | 1 | 0% | 1,737 | 659 | -62% | 0 | 0 | — |
case-12 | fail→pass | 12,027 | 2,659 | -78% | 1 | 1 | 0% | 2,111 | 731 | -65% | 0 | 0 | — |
case-13 | fail→pass | 16,003 | 3,845 | -76% | 1 | 1 | 0% | 2,475 | 945 | -62% | 0 | 0 | — |
case-14 | fail→pass | 6,075 | 3,004 | -51% | 1 | 1 | 0% | 919 | 694 | -24% | 0 | 0 | — |
case-15 | fail→pass | 12,632 | 2,938 | -77% | 1 | 1 | 0% | 1,931 | 773 | -60% | 0 | 0 | — |
case-16 | pass→pass | 9,252 | 2,006 | -78% | 1 | 1 | 0% | 1,275 | 651 | -49% | 0 | 0 | — |
case-17 | pass→pass | 16,407 | 3,348 | -80% | 1 | 1 | 0% | 2,190 | 829 | -62% | 0 | 0 | — |
case-18 | fail→pass | 11,214 | 2,404 | -79% | 1 | 1 | 0% | 1,632 | 670 | -59% | 0 | 0 | — |
case-19 | fail→pass | 11,960 | 3,109 | -74% | 1 | 1 | 0% | 2,003 | 785 | -61% | 0 | 0 | — |
case-20 | fail→pass | 11,704 | 3,215 | -73% | 1 | 1 | 0% | 1,801 | 828 | -54% | 0 | 0 | — |
case-21 | fail→pass | 7,661 | 2,497 | -67% | 1 | 1 | 0% | 812 | 759 | -7% | 0 | 0 | — |
case-22 | pass→pass | 9,189 | 3,116 | -66% | 1 | 1 | 0% | 1,330 | 950 | -29% | 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. The headline lift of +73 percentage points is the difference between those two pass rates over the 22 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.