Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Default reference pipeline for the new-generation taskKind — discovery → plan → generate → critique with a critique-theater devloop.
.claude/skills/nexu-io-od-new-generation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -92% | 0% |
Spec §10.1 / §23.3.3: when a plugin omits od.pipeline, the daemon falls back to this scenario plugin's pipeline. Replacing the default for an enterprise / vertical edition is now a content edit rather than a code change.
jsonc{ "stages": [ { "id": "discovery", "atoms": ["discovery-question-form"] }, { "id": "plan", "atoms": ["direction-picker", "todo-write"] }, { "id": "generate", "atoms": ["file-write", "live-artifact"] }, { "id": "critique", "atoms": ["critique-theater"], "repeat": true, "until": "critique.score>=4 || iterations>=3" } ] }
The discovery stage gives the agent a clean surface for clarifying questions; the plan stage commits to a TodoWrite-backed plan; the generate stage produces the artifact; the critique stage devloops until the score converges or the iteration ceiling is hit.
A plugin can override or extend by declaring its own od.pipeline. The default applies only when the plugin's pipeline is absent.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,377 | 7,486 | -40% | 1 | 1 | 0% | 2,426 | 1,359 | -44% | 0 | 0 | — |
case-02 | fail→pass | 26,158 | 7,039 | -73% | 1 | 1 | 0% | 4,175 | 1,392 | -67% | 0 | 0 | — |
case-03 | fail→pass | 19,184 | 6,583 | -66% | 1 | 1 | 0% | 2,912 | 1,267 | -56% | 0 | 0 | — |
case-04 | fail→pass | 16,905 | 3,467 | -79% | 1 | 1 | 0% | 2,174 | 908 | -58% | 0 | 0 | — |
case-05 | fail→pass | 46,917 | 2,033 | -96% | 1 | 1 | 0% | 8,218 | 655 | -92% | 0 | 0 | — |
case-06 | fail→pass | 46,603 | 1,787 | -96% | 1 | 1 | 0% | 6,510 | 520 | -92% | 0 | 0 | — |
case-07 | fail→fail | 16,779 | 2,066 | -88% | 1 | 1 | 0% | 2,485 | 581 | -77% | 0 | 0 | — |
case-08 | fail→fail | 14,369 | 2,385 | -83% | 1 | 1 | 0% | 1,897 | 542 | -71% | 0 | 0 | — |
case-09 | fail→pass | 20,775 | 2,190 | -89% | 1 | 1 | 0% | 3,503 | 601 | -83% | 0 | 0 | — |
case-10 | fail→pass | 32,649 | 1,593 | -95% | 1 | 1 | 0% | 4,443 | 533 | -88% | 0 | 0 | — |
case-11 | pass→pass | 10,769 | 2,670 | -75% | 1 | 1 | 0% | 1,564 | 647 | -59% | 0 | 0 | — |
case-12 | fail→pass | 11,000 | 2,322 | -79% | 1 | 1 | 0% | 1,800 | 629 | -65% | 0 | 0 | — |
case-13 | fail→pass | 26,089 | 1,759 | -93% | 1 | 1 | 0% | 3,547 | 559 | -84% | 0 | 0 | — |
case-14 | fail→pass | 15,160 | 3,949 | -74% | 1 | 1 | 0% | 1,809 | 768 | -58% | 0 | 0 | — |
case-15 | fail→pass | 24,398 | 1,687 | -93% | 1 | 1 | 0% | 1,706 | 544 | -68% | 0 | 0 | — |
case-16 | fail→pass | 11,054 | 3,627 | -67% | 1 | 1 | 0% | 1,679 | 743 | -56% | 0 | 0 | — |
case-17 | pass→pass | 8,269 | 2,492 | -70% | 1 | 1 | 0% | 1,255 | 511 | -59% | 0 | 0 | — |
case-18 | pass→pass | 32,077 | 2,561 | -92% | 1 | 1 | 0% | 5,729 | 664 | -88% | 0 | 0 | — |
case-19 | fail→pass | 10,539 | 2,063 | -80% | 1 | 1 | 0% | 1,677 | 629 | -62% | 0 | 0 | — |
case-20 | fail→fail | 10,916 | 4,745 | -57% | 1 | 1 | 0% | 1,532 | 1,221 | -20% | 0 | 0 | — |
case-21 | pass→pass | 7,028 | 4,023 | -43% | 1 | 1 | 0% | 1,358 | 992 | -27% | 0 | 0 | — |
case-22 | pass→fail | 20,858 | 11,496 | -45% | 1 | 1 | 0% | 2,837 | 2,391 | -16% | 0 | 0 | — |
case-23 | pass→pass | 15,466 | 2,225 | -86% | 1 | 1 | 0% | 2,660 | 660 | -75% | 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. The headline lift of +57 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.