Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Wenn es um Beihilfe Spezialhilfsmittel — Hoergeraete Cochlea-Implantat Sehhilfen in Beamtenrecht geht: rechnet Schwellen, Beträge, Varianten und Kontrollannahmen durch; liefert eine Berechnungstabelle mit Schwellen, Annahmen und Kontrollfragen.
.claude/skills/klotzkette-beihilfe-implantatfaehige-hoergeraete/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 44% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 54% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 43% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | 39% | 0% |
Skill für Beihilfeberechtigte, denen die Beihilfestelle bei der Erstattung hochwertiger Hilfsmittel die volle Kostenerstattung verweigert hat. Anwendung typischerweise bei Hoergeraeten oberhalb des Festbetrags und bei Cochlea-Implantat-Folgekosten.
beamtenrecht/references/QUELLEN.md; keine BeckRS-, juris-, Kommentar- oder Aufsatz-Blindzitate.Mandant beidseitige Schwerhoerigkeit; HNO empfiehlt Geraet mit Bluetooth und Spezialakustik für Berufstaetigkeit als Richter (Verhandlung in großen Saelen). Kostenvoranschlag 5.800 Euro, Festbetrag 1.500 Euro. Skill liefert Widerspruch mit Begruendung der Mehrkosten als medizinisch notwendig.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 30,664 | 33,863 | +10% | 1 | 1 | 0% | 3,311 | 5,434 | +64% | 0 | 0 | — |
case-02 | pass→pass | 25,539 | 36,856 | +44% | 1 | 1 | 0% | 3,410 | 4,877 | +43% | 0 | 0 | — |
case-03 | fail→pass | 26,407 | 28,561 | +8% | 1 | 1 | 0% | 2,888 | 4,558 | +58% | 0 | 0 | — |
case-04 | fail→pass | 16,447 | 14,818 | -10% | 1 | 1 | 0% | 2,263 | 3,269 | +44% | 0 | 0 | — |
case-05 | fail→fail | 15,291 | 24,692 | +61% | 1 | 1 | 0% | 2,597 | 4,862 | +87% | 0 | 0 | — |
case-06 | fail→fail | 15,591 | 18,837 | +21% | 1 | 1 | 0% | 2,281 | 3,775 | +65% | 0 | 0 | — |
case-07 | pass→pass | 20,948 | 15,029 | -28% | 1 | 1 | 0% | 2,708 | 3,776 | +39% | 0 | 0 | — |
case-08 | pass→pass | 22,198 | 24,480 | +10% | 1 | 1 | 0% | 2,921 | 4,479 | +53% | 0 | 0 | — |
case-09 | fail→pass | 19,406 | 22,220 | +15% | 1 | 1 | 0% | 2,819 | 4,347 | +54% | 0 | 0 | — |
case-10 | pass→pass | 23,405 | 19,695 | -16% | 1 | 1 | 0% | 3,120 | 4,128 | +32% | 0 | 0 | — |
case-11 | fail→fail | 21,058 | 20,183 | -4% | 1 | 1 | 0% | 2,937 | 4,067 | +38% | 0 | 0 | — |
case-12 | pass→pass | 13,962 | 25,470 | +82% | 1 | 1 | 0% | 2,127 | 4,118 | +94% | 0 | 0 | — |
case-13 | pass→pass | 19,235 | 21,071 | +10% | 1 | 1 | 0% | 2,958 | 4,480 | +51% | 0 | 0 | — |
case-14 | pass→pass | 22,232 | 26,097 | +17% | 1 | 1 | 0% | 3,443 | 4,750 | +38% | 0 | 0 | — |
case-15 | pass→pass | 11,693 | 13,379 | +14% | 1 | 1 | 0% | 2,079 | 3,675 | +77% | 0 | 0 | — |
case-16 | pass→pass | 15,651 | 17,070 | +9% | 1 | 1 | 0% | 2,343 | 4,031 | +72% | 0 | 0 | — |
case-17 | pass→pass | 15,597 | 17,521 | +12% | 1 | 1 | 0% | 2,175 | 3,983 | +83% | 0 | 0 | — |
case-18 | pass→pass | 31,922 | 26,181 | -18% | 1 | 1 | 0% | 3,233 | 4,713 | +46% | 0 | 0 | — |
case-19 | pass→pass | 15,623 | 15,472 | -1% | 1 | 1 | 0% | 2,528 | 3,952 | +56% | 0 | 0 | — |
case-20 | pass→pass | 21,581 | 33,521 | +55% | 1 | 1 | 0% | 3,286 | 5,574 | +70% | 0 | 0 | — |
case-21 | pass→pass | 22,147 | 25,116 | +13% | 1 | 1 | 0% | 3,560 | 5,430 | +53% | 0 | 0 | — |
case-22 | pass→pass | 17,820 | 22,964 | +29% | 1 | 1 | 0% | 2,709 | 4,907 | +81% | 0 | 0 | — |
case-23 | pass→pass | 14,844 | 16,260 | +10% | 1 | 1 | 0% | 2,539 | 3,867 | +52% | 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 +13 percentage points is the difference between those two pass rates over the 23 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.