Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →FINRA Broker-Dealer Cybersecurity Guidance expert. Stub-depth framework plugin that routes to the SCF crosswalk. Level up by adding framework-specific context, assessment workflow, and evidence patterns.
.claude/skills/grcengclub-us-finra-expert/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -48% | 0% |
Stub-depth expertise for FINRA Cybersecurity Rules (builds on SEC Reg S-P and SEC 17a-4). This plugin is scaffolded from the SCF crosswalk (17 SCF controls map to 39 framework controls) and defers to /grc-engineer:gap-assessment for the actual compliance check.
usa-federal-sro-finraTODO: replace with framework-specific overview. Minimum sections for Reference-depth upgrade:
All commands in this plugin route through /grc-engineer:gap-assessment with framework ID usa-federal-sro-finra. Reference-depth plugins add:
evidence-checklist — framework-native evidence by control familyscope — applicability determination for the organizationFull-depth plugins add framework-specific workflow commands (examples in sibling plugins like soc2, fedramp-rev5, pci-dss).
See the Framework Plugin Guide for the Stub → Reference → Full progression checklist.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 20,022 | 16,439 | -18% | 1 | 1 | 0% | 3,272 | 2,578 | -21% | 0 | 0 | — |
case-02 | fail→fail | 20,539 | 18,418 | -10% | 1 | 1 | 0% | 3,436 | 3,168 | -8% | 0 | 0 | — |
case-03 | fail→fail | 22,285 | 21,284 | -4% | 1 | 1 | 0% | 3,319 | 3,705 | +12% | 0 | 0 | — |
case-04 | fail→pass | 7,512 | 2,480 | -67% | 1 | 1 | 0% | 1,038 | 714 | -31% | 0 | 0 | — |
case-05 | pass→pass | 6,911 | 7,976 | +15% | 1 | 1 | 0% | 1,105 | 1,697 | +54% | 0 | 0 | — |
case-06 | pass→pass | 7,257 | 6,842 | -6% | 1 | 1 | 0% | 1,145 | 1,384 | +21% | 0 | 0 | — |
case-07 | fail→pass | 9,297 | 3,181 | -66% | 1 | 1 | 0% | 1,437 | 717 | -50% | 0 | 0 | — |
case-08 | fail→pass | 11,790 | 2,999 | -75% | 1 | 1 | 0% | 1,739 | 657 | -62% | 0 | 0 | — |
case-09 | pass→pass | 2,318 | 2,608 | +13% | 1 | 1 | 0% | 357 | 776 | +117% | 0 | 0 | — |
case-10 | fail→pass | 8,340 | 2,926 | -65% | 1 | 1 | 0% | 1,205 | 628 | -48% | 0 | 0 | — |
case-11 | pass→pass | 5,768 | 3,305 | -43% | 1 | 1 | 0% | 765 | 624 | -18% | 0 | 0 | — |
case-12 | fail→pass | 7,568 | 3,249 | -57% | 1 | 1 | 0% | 944 | 756 | -20% | 0 | 0 | — |
case-13 | fail→pass | 7,785 | 3,439 | -56% | 1 | 1 | 0% | 1,029 | 679 | -34% | 0 | 0 | — |
case-14 | fail→pass | 26,391 | 3,583 | -86% | 1 | 1 | 0% | 3,954 | 701 | -82% | 0 | 0 | — |
case-15 | fail→pass | 10,722 | 2,791 | -74% | 1 | 1 | 0% | 1,402 | 681 | -51% | 0 | 0 | — |
case-16 | fail→pass | 5,563 | 3,324 | -40% | 1 | 1 | 0% | 843 | 638 | -24% | 0 | 0 | — |
case-17 | fail→pass | 5,080 | 2,547 | -50% | 1 | 1 | 0% | 721 | 746 | +3% | 0 | 0 | — |
case-18 | pass→pass | 9,575 | 3,028 | -68% | 1 | 1 | 0% | 1,404 | 800 | -43% | 0 | 0 | — |
case-19 | fail→pass | 11,129 | 1,641 | -85% | 1 | 1 | 0% | 1,649 | 589 | -64% | 0 | 0 | — |
case-20 | pass→pass | 16,444 | 12,015 | -27% | 1 | 1 | 0% | 2,490 | 2,266 | -9% | 0 | 0 | — |
case-21 | pass→pass | 14,870 | 12,790 | -14% | 1 | 1 | 0% | 2,233 | 2,332 | +4% | 0 | 0 | — |
case-22 | fail→pass | 11,823 | 8,327 | -30% | 1 | 1 | 0% | 1,789 | 1,644 | -8% | 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. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 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.