Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when choosing native or multi-LLM handling for init, review, or security requests
.claude/skills/nyldn-skill-native-escalation-routing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 18% | 0% |
Use this skill when the user asks for repository initialization, code review, or security review and it is not yet clear whether Claude-native behavior is sufficient or whether Octopus escalation is warranted.
Claude-native first:
/init/review/security-reviewOctopus for escalation:
/octo:review/octo:security/octo:debate/octo:multiPrefer Claude-native behavior when all of the following are true:
Examples:
Escalate when the user asks for or clearly benefits from:
Examples:
/octo:review or /octo:security, treat that as an escalation request.Use wording like:
> Claude-native first, Octopus for escalation. Use Claude-native /review or /security-review for ordinary requests. Use Octopus when you want multiple model opinions, adversarial review, or stricter multi-LLM workflows.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 14,355 | 4,760 | -67% | 1 | 1 | 0% | 2,414 | 1,219 | -50% | 0 | 0 | — |
case-02 | fail→pass | 8,065 | 3,937 | -51% | 1 | 1 | 0% | 1,213 | 1,074 | -11% | 0 | 0 | — |
case-03 | fail→pass | 13,697 | 5,880 | -57% | 1 | 1 | 0% | 2,118 | 1,404 | -34% | 0 | 0 | — |
case-04 | fail→pass | 7,059 | 3,612 | -49% | 1 | 1 | 0% | 1,002 | 1,043 | +4% | 0 | 0 | — |
case-05 | fail→pass | 7,330 | 8,763 | +20% | 1 | 1 | 0% | 995 | 1,172 | +18% | 0 | 0 | — |
case-06 | fail→pass | 7,914 | 4,449 | -44% | 1 | 1 | 0% | 1,202 | 967 | -20% | 0 | 0 | — |
case-07 | pass→pass | 13,013 | 6,382 | -51% | 1 | 1 | 0% | 2,003 | 867 | -57% | 0 | 0 | — |
case-08 | fail→pass | 20,309 | 11,278 | -44% | 1 | 1 | 0% | 1,723 | 1,118 | -35% | 0 | 0 | — |
case-09 | fail→pass | 15,114 | 3,797 | -75% | 1 | 1 | 0% | 2,440 | 1,010 | -59% | 0 | 0 | — |
case-10 | fail→pass | 8,377 | 4,755 | -43% | 1 | 1 | 0% | 1,333 | 1,143 | -14% | 0 | 0 | — |
case-11 | fail→pass | 7,753 | 3,085 | -60% | 1 | 1 | 0% | 1,156 | 909 | -21% | 0 | 0 | — |
case-12 | fail→pass | 5,323 | 2,641 | -50% | 1 | 1 | 0% | 751 | 861 | +15% | 0 | 0 | — |
case-13 | fail→pass | 12,305 | 4,807 | -61% | 1 | 1 | 0% | 1,970 | 1,191 | -40% | 0 | 0 | — |
case-14 | fail→pass | 8,363 | 5,570 | -33% | 1 | 1 | 0% | 1,272 | 1,228 | -3% | 0 | 0 | — |
case-15 | fail→pass | 7,011 | 2,904 | -59% | 1 | 1 | 0% | 1,231 | 875 | -29% | 0 | 0 | — |
case-16 | fail→pass | 15,700 | 12,382 | -21% | 1 | 1 | 0% | 2,025 | 1,131 | -44% | 0 | 0 | — |
case-21 | fail→pass | 3,691 | 3,269 | -11% | 1 | 1 | 0% | 499 | 931 | +87% | 0 | 0 | — |
case-17 | pass→pass | 4,671 | 10,930 | +134% | 1 | 1 | 0% | 654 | 926 | +42% | 0 | 0 | — |
case-18 | fail→pass | 14,843 | 4,247 | -71% | 1 | 1 | 0% | 2,235 | 1,201 | -46% | 0 | 0 | — |
case-19 | fail→pass | 6,303 | 9,548 | +51% | 1 | 1 | 0% | 1,137 | 966 | -15% | 0 | 0 | — |
case-20 | pass→pass | 5,817 | 3,275 | -44% | 1 | 1 | 0% | 845 | 933 | +10% | 0 | 0 | — |
case-22 | fail→pass | 12,215 | 4,879 | -60% | 1 | 1 | 0% | 1,946 | 1,221 | -37% | 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 +86 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/11/2026 | +70% |
Other measured skills in the registry, with their headline benchmark lift.