Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generates focused assessor interview questions from a technology stack and maps them to compliance frameworks with practical CLI/API evidence hints.
.claude/skills/grcengclub-interview-question-generator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -72% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -57% | 0% |
Use this skill when preparing audit or assessment interview questions for a client environment. It helps avoid broad screenshot requests by producing targeted questions, expected evidence, and programmatic collection hints.
Generate markdown questions:
bashnode plugins/grc-engineer/scripts/generate-interview-questions.js "AWS IAM users" SOC2,NIST-800-53,PCI-DSS
Generate JSON for downstream tooling:
bashnode plugins/grc-engineer/scripts/generate-interview-questions.js "GitHub branch protection" SOC2 --format=json
Limit output for a short interview block:
bashnode plugins/grc-engineer/scripts/generate-interview-questions.js "Okta MFA" ISO27001 --limit=5
If no stack-specific profile matches, the generator falls back to a generic assessment profile.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 22,026 | 20,495 | -7% | 1 | 1 | 0% | 3,595 | 3,540 | -2% | 0 | 0 | — |
case-02 | fail→fail | 19,255 | 15,880 | -18% | 1 | 1 | 0% | 3,553 | 3,276 | -8% | 0 | 0 | — |
case-03 | fail→fail | 11,304 | 10,360 | -8% | 1 | 1 | 0% | 1,675 | 2,048 | +22% | 0 | 0 | — |
case-04 | fail→pass | 12,421 | 3,516 | -72% | 1 | 1 | 0% | 1,875 | 841 | -55% | 0 | 0 | — |
case-05 | fail→fail | 6,827 | 30,120 | +341% | 1 | 1 | 0% | 1,008 | 2,080 | +106% | 0 | 0 | — |
case-06 | fail→pass | 15,216 | 3,117 | -80% | 1 | 1 | 0% | 2,401 | 791 | -67% | 0 | 0 | — |
case-07 | fail→pass | 12,288 | 2,545 | -79% | 1 | 1 | 0% | 1,870 | 705 | -62% | 0 | 0 | — |
case-08 | fail→fail | 19,249 | 5,746 | -70% | 1 | 1 | 0% | 2,917 | 1,245 | -57% | 0 | 0 | — |
case-09 | fail→pass | 15,678 | 2,526 | -84% | 1 | 1 | 0% | 2,525 | 695 | -72% | 0 | 0 | — |
case-10 | fail→pass | 15,549 | 4,553 | -71% | 1 | 1 | 0% | 2,406 | 1,031 | -57% | 0 | 0 | — |
case-11 | fail→pass | 12,456 | 3,071 | -75% | 1 | 1 | 0% | 1,995 | 818 | -59% | 0 | 0 | — |
case-12 | fail→pass | 6,976 | 1,739 | -75% | 1 | 1 | 0% | 1,076 | 568 | -47% | 0 | 0 | — |
case-13 | fail→pass | 18,375 | 1,957 | -89% | 1 | 1 | 0% | 3,085 | 576 | -81% | 0 | 0 | — |
case-14 | fail→pass | 15,497 | 2,860 | -82% | 1 | 1 | 0% | 2,334 | 759 | -67% | 0 | 0 | — |
case-15 | fail→pass | 15,095 | 2,680 | -82% | 1 | 1 | 0% | 2,240 | 739 | -67% | 0 | 0 | — |
case-16 | fail→pass | 9,072 | 2,100 | -77% | 1 | 1 | 0% | 1,347 | 622 | -54% | 0 | 0 | — |
case-17 | fail→pass | 10,864 | 2,390 | -78% | 1 | 1 | 0% | 1,721 | 605 | -65% | 0 | 0 | — |
case-18 | fail→pass | 12,214 | 2,747 | -78% | 1 | 1 | 0% | 1,953 | 765 | -61% | 0 | 0 | — |
case-19 | fail→pass | 12,328 | 2,505 | -80% | 1 | 1 | 0% | 2,099 | 675 | -68% | 0 | 0 | — |
case-20 | pass→pass | 4,356 | 5,546 | +27% | 1 | 1 | 0% | 783 | 1,290 | +65% | 0 | 0 | — |
case-21 | pass→pass | 12,349 | 12,476 | +1% | 1 | 1 | 0% | 1,884 | 2,291 | +22% | 0 | 0 | — |
case-22 | pass→pass | 4,341 | 3,673 | -15% | 1 | 1 | 0% | 694 | 917 | +32% | 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 +64 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.