Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Converts natural language compliance requirements into executable policies (OPA Rego, AWS Config Rules, Sentinel, Terraform). Standardizes governance by making it part of the build process.
.claude/skills/grcengclub-policy-as-code-generator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -37% | 0% |
Converts natural language compliance requirements into executable policy code. Generates OPA Rego, AWS Config Rules, Sentinel policies, or Terraform modules.
Generate OPA Rego policy:
bashnode plugins/grc-engineer/scripts/generate-policy.js "Ensure no S3 buckets are public and all must have a 'Department' tag" rego
Generate AWS Config Rule:
bashnode plugins/grc-engineer/scripts/generate-policy.js "All EC2 instances must have encryption enabled" aws-config
Generate Sentinel policy:
bashnode plugins/grc-engineer/scripts/generate-policy.js "Terraform plans must not create resources without required tags" sentinel
Generated policies include:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 22,081 | 34,597 | +57% | 1 | 1 | 0% | 4,198 | 4,000 | -5% | 0 | 0 | — |
case-02 | fail→fail | 17,444 | 13,508 | -23% | 1 | 1 | 0% | 2,967 | 2,808 | -5% | 0 | 0 | — |
case-03 | fail→pass | 11,902 | 3,131 | -74% | 1 | 1 | 0% | 2,026 | 844 | -58% | 0 | 0 | — |
case-04 | fail→pass | 10,741 | 2,770 | -74% | 1 | 1 | 0% | 1,648 | 823 | -50% | 0 | 0 | — |
case-05 | fail→pass | 11,489 | 2,227 | -81% | 1 | 1 | 0% | 1,911 | 746 | -61% | 0 | 0 | — |
case-06 | fail→pass | 7,898 | 2,374 | -70% | 1 | 1 | 0% | 1,119 | 745 | -33% | 0 | 0 | — |
case-07 | fail→pass | 7,863 | 2,400 | -69% | 1 | 1 | 0% | 1,177 | 744 | -37% | 0 | 0 | — |
case-08 | fail→pass | 18,997 | 2,097 | -89% | 1 | 1 | 0% | 1,266 | 706 | -44% | 0 | 0 | — |
case-09 | fail→pass | 14,538 | 2,463 | -83% | 1 | 1 | 0% | 2,271 | 740 | -67% | 0 | 0 | — |
case-10 | fail→pass | 5,174 | 2,017 | -61% | 1 | 1 | 0% | 889 | 660 | -26% | 0 | 0 | — |
case-11 | fail→pass | 8,339 | 2,183 | -74% | 1 | 1 | 0% | 1,153 | 733 | -36% | 0 | 0 | — |
case-12 | fail→pass | 20,345 | 2,409 | -88% | 1 | 1 | 0% | 3,158 | 743 | -76% | 0 | 0 | — |
case-13 | fail→fail | 14,171 | 2,634 | -81% | 1 | 1 | 0% | 2,163 | 749 | -65% | 0 | 0 | — |
case-14 | fail→pass | 11,673 | 3,822 | -67% | 1 | 1 | 0% | 1,687 | 958 | -43% | 0 | 0 | — |
case-15 | fail→pass | 11,177 | 1,983 | -82% | 1 | 1 | 0% | 1,761 | 670 | -62% | 0 | 0 | — |
case-16 | fail→pass | 27,430 | 2,686 | -90% | 1 | 1 | 0% | 1,898 | 676 | -64% | 0 | 0 | — |
case-17 | fail→pass | 9,655 | 2,317 | -76% | 1 | 1 | 0% | 1,624 | 757 | -53% | 0 | 0 | — |
case-18 | fail→pass | 13,585 | 2,490 | -82% | 1 | 1 | 0% | 2,222 | 729 | -67% | 0 | 0 | — |
case-19 | pass→pass | 8,957 | 2,751 | -69% | 1 | 1 | 0% | 1,354 | 811 | -40% | 0 | 0 | — |
case-20 | pass→pass | 11,405 | 8,427 | -26% | 1 | 1 | 0% | 1,864 | 1,810 | -3% | 0 | 0 | — |
case-21 | pass→pass | 4,180 | 3,597 | -14% | 1 | 1 | 0% | 758 | 1,071 | +41% | 0 | 0 | — |
case-22 | pass→pass | 5,604 | 4,375 | -22% | 1 | 1 | 0% | 787 | 1,053 | +34% | 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 +68 percentage points is the difference between those two pass rates over the 21 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.