Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Staff a task only from Agentlas agents registered on this machine.
.claude/skills/agentlas-ai-hep-local/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 106% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -82% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -62% | 0% |
Update fallback: 자동 업데이트가 안 되면 hephaestus update를 한 번 실행하세요. 업데이트하지 않아도 현재 버전 명령은 그대로 동작합니다.
Raw request: $ARGUMENTS
Use MCP server hephaestus-network and exact sourceScope: "local". Author a redacted agentlas.workforce-work-order.v1, call workforce.search_candidates with {workOrder, sourceScope: "local"} and keep the response as federationResult. Author the final agentlas.workforce-selection.v1 yourself, call workforce.validate_selection with {workOrder, selection}, keep its response as federatedSelection, then call workforce.prepare_execution with {workOrder, selection, federatedSelection, projectDir}. Require every row to retain source local plus its exact package/content/runtime/permission/context identity.
Run planner/manager, selected workers, synthesis, and verifier as distinct invocations with artifact handoffs and preserve Team graphs. If Core or the registered Local inventory is unavailable, report source_unavailable. Never search Cloud or Hub, accept a deterministic picker, or treat a prepared bundle as execution proof.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 10,236 | 23,086 | +126% | 1 | 1 | 0% | 251 | 888 | +254% | 0 | 0 | — |
case-02 | fail→fail | 33,758 | 44,453 | +32% | 1 | 1 | 0% | 3,367 | 2,620 | -22% | 0 | 0 | — |
case-03 | fail→fail | 30,021 | 18,133 | -40% | 1 | 1 | 0% | 5,573 | 962 | -83% | 0 | 0 | — |
case-04 | fail→fail | 11,293 | 17,592 | +56% | 1 | 1 | 0% | 1,718 | 1,017 | -41% | 0 | 0 | — |
case-05 | fail→pass | 22,827 | 10,956 | -52% | 1 | 1 | 0% | 1,065 | 2,198 | +106% | 0 | 0 | — |
case-06 | fail→pass | 42,056 | 2,968 | -93% | 1 | 1 | 0% | 1,342 | 622 | -54% | 0 | 0 | — |
case-07 | fail→pass | 24,754 | 3,507 | -86% | 1 | 1 | 0% | 4,245 | 780 | -82% | 0 | 0 | — |
case-08 | fail→pass | 10,711 | 9,489 | -11% | 1 | 1 | 0% | 1,648 | 1,813 | +10% | 0 | 0 | — |
case-09 | pass→pass | 12,549 | 10,515 | -16% | 1 | 1 | 0% | 2,186 | 1,544 | -29% | 0 | 0 | — |
case-10 | fail→fail | 30,775 | 21,271 | -31% | 1 | 1 | 0% | 2,579 | 996 | -61% | 0 | 0 | — |
case-11 | fail→pass | 16,854 | 5,392 | -68% | 1 | 1 | 0% | 2,580 | 986 | -62% | 0 | 0 | — |
case-12 | pass→fail | 8,476 | 26,938 | +218% | 1 | 1 | 0% | 1,074 | 1,476 | +37% | 0 | 0 | — |
case-13 | fail→pass | 11,340 | 10,730 | -5% | 1 | 1 | 0% | 1,659 | 2,114 | +27% | 0 | 0 | — |
case-14 | fail→pass | 9,091 | 3,286 | -64% | 1 | 1 | 0% | 1,312 | 520 | -60% | 0 | 0 | — |
case-15 | fail→pass | 17,889 | 4,624 | -74% | 1 | 1 | 0% | 2,679 | 935 | -65% | 0 | 0 | — |
case-16 | fail→pass | 12,672 | 7,266 | -43% | 1 | 1 | 0% | 1,951 | 1,533 | -21% | 0 | 0 | — |
case-17 | fail→pass | 25,402 | 18,106 | -29% | 1 | 1 | 0% | 1,995 | 2,997 | +50% | 0 | 0 | — |
case-18 | fail→pass | 13,387 | 16,362 | +22% | 1 | 1 | 0% | 2,013 | 1,788 | -11% | 0 | 0 | — |
case-19 | fail→pass | 16,231 | 5,291 | -67% | 1 | 1 | 0% | 2,073 | 735 | -65% | 0 | 0 | — |
case-20 | fail→fail | 19,789 | 18,816 | -5% | 1 | 1 | 0% | 2,851 | 2,321 | -19% | 0 | 0 | — |
case-21 | fail→pass | 14,985 | 16,125 | +8% | 1 | 1 | 0% | 2,115 | 2,420 | +14% | 0 | 0 | — |
case-22 | pass→fail | 12,903 | 7,942 | -38% | 1 | 1 | 0% | 1,908 | 785 | -59% | 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 13 counted toward the lift figure. The other 9 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 +50 percentage points is the difference between those two pass rates over the 13 comparable cases. 2 cases got worse with the skill loaded, and they are 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.