Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Open Token Harbor, report stored Sail Power and harbor progress, or help the user configure local Codex token tracking.
.claude/skills/hashgraph-online-token-harbor/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -81% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -73% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -61% | 0% |
Use the bundled token-harbor MCP tools whenever the user asks to open the harbor, check Sail Power, inspect port progress, or troubleshoot the companion game.
open_harbor when the user wants to play or see the harbor.harbor_status for a concise progress summary when a full game view is unnecessary.PLUGIN_DATA.otel.log_user_prompt = false when helping with configuration.The game may be active while Codex works. When a task needs approval or finishes, the harbor pauses its active voyage and highlights the task status so the user can return to Codex immediately.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 13,946 | 14,677 | +5% | 1 | 1 | 0% | 979 | 553 | -44% | 0 | 0 | — |
case-01 | fail→fail | 7,455 | 14,746 | +98% | 1 | 1 | 0% | 362 | 546 | +51% | 0 | 0 | — |
case-02 | fail→fail | 10,533 | 13,551 | +29% | 1 | 1 | 0% | 678 | 486 | -28% | 0 | 0 | — |
case-04 | fail→fail | 9,394 | 12,969 | +38% | 1 | 1 | 0% | 616 | 418 | -32% | 0 | 0 | — |
case-05 | fail→fail | 12,259 | 10,479 | -15% | 1 | 1 | 0% | 1,168 | 635 | -46% | 0 | 0 | — |
case-06 | fail→fail | 11,197 | 4,183 | -63% | 1 | 1 | 0% | 1,697 | 781 | -54% | 0 | 0 | — |
case-07 | fail→pass | 19,340 | 14,939 | -23% | 1 | 1 | 0% | 2,537 | 1,637 | -35% | 0 | 0 | — |
case-08 | fail→pass | 24,249 | 6,981 | -71% | 1 | 1 | 0% | 2,645 | 493 | -81% | 0 | 0 | — |
case-09 | pass→pass | 13,833 | 8,668 | -37% | 1 | 1 | 0% | 1,817 | 830 | -54% | 0 | 0 | — |
case-10 | fail→pass | 13,803 | 2,859 | -79% | 1 | 1 | 0% | 1,219 | 656 | -46% | 0 | 0 | — |
case-11 | pass→pass | 20,404 | 8,689 | -57% | 1 | 1 | 0% | 1,741 | 565 | -68% | 0 | 0 | — |
case-12 | fail→pass | 19,500 | 3,465 | -82% | 1 | 1 | 0% | 2,343 | 640 | -73% | 0 | 0 | — |
case-13 | fail→pass | 10,886 | 3,923 | -64% | 1 | 1 | 0% | 1,761 | 688 | -61% | 0 | 0 | — |
case-14 | pass→pass | 12,630 | 7,212 | -43% | 1 | 1 | 0% | 1,768 | 540 | -69% | 0 | 0 | — |
case-15 | fail→fail | 8,150 | 9,196 | +13% | 1 | 1 | 0% | 1,068 | 398 | -63% | 0 | 0 | — |
case-16 | fail→fail | 4,884 | 8,205 | +68% | 1 | 1 | 0% | 808 | 438 | -46% | 0 | 0 | — |
case-17 | fail→pass | 19,165 | 3,383 | -82% | 1 | 1 | 0% | 1,946 | 677 | -65% | 0 | 0 | — |
case-18 | fail→pass | 24,146 | 8,002 | -67% | 1 | 1 | 0% | 2,765 | 518 | -81% | 0 | 0 | — |
case-19 | fail→pass | 31,296 | 6,684 | -79% | 1 | 1 | 0% | 4,216 | 476 | -89% | 0 | 0 | — |
case-20 | pass→pass | 28,873 | 17,244 | -40% | 1 | 1 | 0% | 3,808 | 2,858 | -25% | 0 | 0 | — |
case-21 | pass→pass | 14,315 | 14,115 | -1% | 1 | 1 | 0% | 1,687 | 1,533 | -9% | 0 | 0 | — |
case-22 | pass→pass | 19,743 | 12,581 | -36% | 1 | 1 | 0% | 2,895 | 1,709 | -41% | 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 16 counted toward the lift figure. The other 6 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 16 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.