Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Wenn es um Pfandbrief Spezialdeckung in Bank-Rechtsabteilung geht: erstellt den passenden Entwurf aus Sachverhalt, Norm, Beweis und Antrag; liefert einen verwertbaren Entwurf mit Anträgen, Begründung und Anlagenlogik.
.claude/skills/klotzkette-pfandbriefbank-spezialdeckung/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-16 | ✗→✓ | ▲ Improved | 121% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 96% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 136% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 141% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 104% | 0% |
Frage nur nach, wenn ohne Antwort ein falscher nächster Schritt droht. Andernfalls mit Annahmen arbeiten und sie sichtbar markieren.
| Punkt | Klärung | | --- | --- | | Ergebnisbedarf | Vermerk, Freigabe, BaFin-Anfrage, Vertrag, Vorstandsvorlage oder Prozessstrategie | | Rechtsregime | KWG, ZAG, WpHG, WpIG, eWpG, MiCAR, DORA, GwG, BGB, HGB, AktG, SEPA-/EU-Regime oder Entwurf | | Risiko | Aufsicht, Bußgeld, Zivilhaftung, Organhaftung, Kundenstreit, AML, Datenschutz, IT oder Reputation | | Frist | Anzeige, Launch, Antwort, Rückgabe, Register, Gremium oder Verjährung | | Entscheidung | Go, Go mit Auflagen, Stop, BaFin-Vorabklärung oder externe Spezialprüfung |
Arbeite in dieser Reihenfolge:
| Frage | Beleg | Fehlt | Owner | Wirkung | | --- | --- | --- | --- | --- | | Wer erbringt welche Leistung? | Vertrag, Produktbild, Prozess | ... | Legal/Produkt | Regimewahl | | Fließen Kundengelder oder Kryptowerte? | Flow-of-Funds, Wallet-/Kontoauszug | ... | Operations/Risk | Erlaubnis/Haftung | | Gibt es Aufsichtskontakt? | Schreiben, Ticket, Portalnachweis | ... | Legal/Compliance | Frist/Strategie | | Sind Kunden betroffen? | AGB, FAQ, Beschwerde, Marketing | ... | Vertrieb/Legal | Transparenz/Haftung |
Erzeuge Deckungscheck, Dokumentenliste, Rechtsrisiken und Gremiennotiz.
Baue das Ergebnis mit diesen Elementen:
bank-rechtsabteilung nennen, insbesondere bafin-kommunikation-und-anhoerung, bankaufsichtsrecht-kwg-marisk-triage, dora-ict-vertraege-vorfall, gwg-aml-kyc-verdachtsmeldung, vorstandsvorlage-gutachten oder produktfreigabe-new-product-process.Vor Ausgabe prüfen:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 37,436 | 36,166 | -3% | 1 | 1 | 0% | 4,965 | 7,230 | +46% | 0 | 0 | — |
case-01 | fail→fail | 35,389 | 45,789 | +29% | 1 | 1 | 0% | 4,794 | 8,494 | +77% | 0 | 0 | — |
case-03 | fail→fail | 34,373 | 43,007 | +25% | 1 | 1 | 0% | 4,739 | 7,928 | +67% | 0 | 0 | — |
case-04 | pass→pass | 21,857 | 31,573 | +44% | 1 | 1 | 0% | 3,463 | 5,894 | +70% | 0 | 0 | — |
case-05 | pass→fail | 18,640 | 34,447 | +85% | 1 | 1 | 0% | 2,976 | 6,071 | +104% | 0 | 0 | — |
case-06 | fail→fail | 15,633 | 31,226 | +100% | 1 | 1 | 0% | 2,510 | 6,226 | +148% | 0 | 0 | — |
case-07 | fail→fail | 17,581 | 25,047 | +42% | 1 | 1 | 0% | 2,147 | 5,316 | +148% | 0 | 0 | — |
case-20 | fail→fail | 21,789 | 35,415 | +63% | 1 | 1 | 0% | 2,682 | 6,490 | +142% | 0 | 0 | — |
case-08 | fail→fail | 26,262 | 43,535 | +66% | 1 | 1 | 0% | 3,744 | 8,609 | +130% | 0 | 0 | — |
case-09 | fail→fail | 26,863 | 32,272 | +20% | 1 | 1 | 0% | 3,279 | 6,302 | +92% | 0 | 0 | — |
case-10 | fail→fail | 26,959 | 37,531 | +39% | 1 | 1 | 0% | 3,527 | 7,331 | +108% | 0 | 0 | — |
case-11 | pass→pass | 28,547 | 32,770 | +15% | 1 | 1 | 0% | 3,118 | 6,602 | +112% | 0 | 0 | — |
case-12 | fail→fail | 23,013 | 30,072 | +31% | 1 | 1 | 0% | 2,469 | 5,917 | +140% | 0 | 0 | — |
case-13 | fail→fail | 17,634 | 29,353 | +66% | 1 | 1 | 0% | 1,966 | 5,771 | +194% | 0 | 0 | — |
case-14 | fail→fail | 22,671 | 30,611 | +35% | 1 | 1 | 0% | 3,348 | 6,631 | +98% | 0 | 0 | — |
case-15 | pass→fail | 28,412 | 31,301 | +10% | 1 | 1 | 0% | 3,800 | 6,477 | +70% | 0 | 0 | — |
case-16 | fail→pass | 22,734 | 32,136 | +41% | 1 | 1 | 0% | 2,875 | 6,343 | +121% | 0 | 0 | — |
case-17 | pass→pass | 24,155 | 44,663 | +85% | 1 | 1 | 0% | 3,001 | 6,377 | +112% | 0 | 0 | — |
case-18 | fail→pass | 29,728 | 38,248 | +29% | 1 | 1 | 0% | 3,745 | 7,348 | +96% | 0 | 0 | — |
case-19 | fail→pass | 22,506 | 31,488 | +40% | 1 | 1 | 0% | 2,545 | 6,007 | +136% | 0 | 0 | — |
case-21 | fail→pass | 21,747 | 33,836 | +56% | 1 | 1 | 0% | 2,549 | 6,138 | +141% | 0 | 0 | — |
case-22 | fail→fail | 19,054 | 38,787 | +104% | 1 | 1 | 0% | 2,188 | 4,980 | +128% | 0 | 0 | — |
case-23 | pass→pass | 33,279 | 52,682 | +58% | 1 | 1 | 0% | 4,685 | 9,161 | +96% | 0 | 0 | — |
case-24 | pass→pass | 29,429 | 32,630 | +11% | 1 | 1 | 0% | 4,191 | 6,589 | +57% | 0 | 0 | — |
case-25 | pass→pass | 26,428 | 30,532 | +16% | 1 | 1 | 0% | 3,654 | 7,148 | +96% | 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. 25 cases were attempted. The headline lift of +8 percentage points is the difference between those two pass rates over the 25 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.