Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Подготовь Gemini deploy/release handoff. EN: prepare Gemini deploy or release handoff.
.claude/skills/bilal140202-ry-deploy/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -16% | 0% |
Prepare and verify deployment, release, or publication work for Gemini adapter changes.
Gemini headless prompts are allowed for smoke/doctor/CI checks only. Persistent orchestration remains cmux visible-terminal-only.
Use when a request includes deploy, release, publish, rollout, server, or production handoff.
Commands, artifact paths, release tags, GitHub run URLs, health check output, and explicit NOT_PROVEN gaps.
Do not invent credentials, servers, release tags, CI proof, or background orchestrators.
Release/deploy state is reproducible from source-of-truth commands and clean artifacts.
Lead with blockers that prevent safe deploy or release.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 8,226 | 7,854 | -5% | 1 | 1 | 0% | 1,521 | 1,657 | +9% | 0 | 0 | — |
case-01 | fail→pass | 10,701 | 9,066 | -15% | 1 | 1 | 0% | 1,931 | 2,025 | +5% | 0 | 0 | — |
case-02 | fail→fail | 8,237 | 4,605 | -44% | 1 | 1 | 0% | 1,480 | 656 | -56% | 0 | 0 | — |
case-04 | fail→pass | 9,062 | 4,531 | -50% | 1 | 1 | 0% | 1,626 | 1,084 | -33% | 0 | 0 | — |
case-05 | pass→pass | 16,514 | 6,817 | -59% | 1 | 1 | 0% | 2,925 | 1,449 | -50% | 0 | 0 | — |
case-06 | fail→pass | 8,840 | 6,665 | -25% | 1 | 1 | 0% | 1,658 | 1,513 | -9% | 0 | 0 | — |
case-07 | pass→pass | 3,923 | 4,993 | +27% | 1 | 1 | 0% | 685 | 1,112 | +62% | 0 | 0 | — |
case-17 | fail→pass | 5,204 | 2,393 | -54% | 1 | 1 | 0% | 849 | 713 | -16% | 0 | 0 | — |
case-08 | pass→pass | 9,588 | 3,759 | -61% | 1 | 1 | 0% | 1,601 | 978 | -39% | 0 | 0 | — |
case-09 | fail→pass | 11,408 | 4,647 | -59% | 1 | 1 | 0% | 1,608 | 1,006 | -37% | 0 | 0 | — |
case-10 | fail→fail | 6,925 | 9,244 | +33% | 1 | 1 | 0% | 1,055 | 1,615 | +53% | 0 | 0 | — |
case-11 | fail→pass | 12,213 | 7,306 | -40% | 1 | 1 | 0% | 1,927 | 1,520 | -21% | 0 | 0 | — |
case-12 | pass→pass | 10,372 | 6,760 | -35% | 1 | 1 | 0% | 1,409 | 1,285 | -9% | 0 | 0 | — |
case-13 | fail→pass | 10,041 | 6,392 | -36% | 1 | 1 | 0% | 1,609 | 1,125 | -30% | 0 | 0 | — |
case-14 | fail→pass | 12,482 | 7,072 | -43% | 1 | 1 | 0% | 1,919 | 1,266 | -34% | 0 | 0 | — |
case-15 | pass→pass | 12,955 | 9,052 | -30% | 1 | 1 | 0% | 2,077 | 1,596 | -23% | 0 | 0 | — |
case-16 | pass→pass | 7,575 | 4,829 | -36% | 1 | 1 | 0% | 1,281 | 1,123 | -12% | 0 | 0 | — |
case-18 | pass→pass | 10,842 | 7,126 | -34% | 1 | 1 | 0% | 1,775 | 1,535 | -14% | 0 | 0 | — |
case-19 | pass→pass | 7,970 | 2,843 | -64% | 1 | 1 | 0% | 1,254 | 781 | -38% | 0 | 0 | — |
case-20 | pass→pass | 15,826 | 13,119 | -17% | 1 | 1 | 0% | 3,263 | 3,146 | -4% | 0 | 0 | — |
case-21 | pass→fail | 12,321 | 12,865 | +4% | 1 | 1 | 0% | 2,429 | 2,486 | +2% | 0 | 0 | — |
case-22 | pass→pass | 5,638 | 4,084 | -28% | 1 | 1 | 0% | 1,162 | 1,015 | -13% | 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 21 counted toward the lift figure. The other 1 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 21 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.