Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Enforces GitHub Actions security and compliance for this monorepo. Use when adding third-party actions, handling secrets, defining permissions, and reviewing CI security trade-offs.
.claude/skills/camunda-ci-security-compliance/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -30% | 0% |
Consult docs/monorepo-docs/ci.md before decisions.
docs/monorepo-docs/ci.md (CI Security section).Never silently drop unapproved but relevant options.
Start with:
yamlpermissions: {}
Add only required scopes.
Vault is mandatory. GitHub Secrets are only for Vault bootstrap values:
VAULT_ADDRVAULT_ROLE_IDVAULT_SECRET_IDcurl | bashreferences/approved-actions.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 6,730 | 7,610 | +13% | 1 | 1 | 0% | 1,354 | 1,910 | +41% | 0 | 0 | — |
case-06 | pass→pass | 11,309 | 8,904 | -21% | 1 | 1 | 0% | 2,268 | 2,135 | -6% | 0 | 0 | — |
case-04 | pass→pass | 11,186 | 6,423 | -43% | 1 | 1 | 0% | 2,179 | 1,525 | -30% | 0 | 0 | — |
case-01 | fail→pass | 15,465 | 10,639 | -31% | 1 | 1 | 0% | 2,996 | 2,320 | -23% | 0 | 0 | — |
case-02 | fail→pass | 9,313 | 9,893 | +6% | 1 | 1 | 0% | 1,600 | 2,145 | +34% | 0 | 0 | — |
case-03 | fail→pass | 14,848 | 6,999 | -53% | 1 | 1 | 0% | 2,603 | 1,697 | -35% | 0 | 0 | — |
case-07 | pass→pass | 7,656 | 5,851 | -24% | 1 | 1 | 0% | 1,503 | 1,347 | -10% | 0 | 0 | — |
case-08 | pass→pass | 10,367 | 4,994 | -52% | 1 | 1 | 0% | 1,919 | 1,271 | -34% | 0 | 0 | — |
case-09 | fail→pass | 9,914 | 6,187 | -38% | 1 | 1 | 0% | 1,736 | 1,478 | -15% | 0 | 0 | — |
case-10 | fail→pass | 6,674 | 2,525 | -62% | 1 | 1 | 0% | 1,133 | 790 | -30% | 0 | 0 | — |
case-11 | fail→pass | 9,968 | 4,652 | -53% | 1 | 1 | 0% | 1,928 | 1,256 | -35% | 0 | 0 | — |
case-12 | fail→pass | 5,839 | 1,752 | -70% | 1 | 1 | 0% | 1,023 | 575 | -44% | 0 | 0 | — |
case-13 | fail→pass | 4,154 | 2,222 | -47% | 1 | 1 | 0% | 663 | 674 | +2% | 0 | 0 | — |
case-14 | fail→pass | 9,730 | 7,694 | -21% | 1 | 1 | 0% | 1,630 | 505 | -69% | 0 | 0 | — |
case-15 | fail→pass | 9,618 | 5,364 | -44% | 1 | 1 | 0% | 1,685 | 1,164 | -31% | 0 | 0 | — |
case-16 | fail→pass | 9,746 | 4,643 | -52% | 1 | 1 | 0% | 1,692 | 1,102 | -35% | 0 | 0 | — |
case-17 | pass→pass | 11,755 | 4,667 | -60% | 1 | 1 | 0% | 2,122 | 1,097 | -48% | 0 | 0 | — |
case-18 | fail→pass | 7,798 | 2,625 | -66% | 1 | 1 | 0% | 1,402 | 828 | -41% | 0 | 0 | — |
case-19 | fail→pass | 6,140 | 3,522 | -43% | 1 | 1 | 0% | 1,043 | 623 | -40% | 0 | 0 | — |
case-20 | fail→pass | 11,135 | 5,325 | -52% | 1 | 1 | 0% | 1,787 | 1,227 | -31% | 0 | 0 | — |
case-21 | fail→pass | 7,214 | 1,848 | -74% | 1 | 1 | 0% | 1,206 | 570 | -53% | 0 | 0 | — |
case-22 | fail→pass | 11,760 | 6,113 | -48% | 1 | 1 | 0% | 2,087 | 1,474 | -29% | 0 | 0 | — |
case-23 | fail→pass | 3,423 | 1,159 | -66% | 1 | 1 | 0% | 658 | 431 | -34% | 0 | 0 | — |
case-24 | fail→pass | 4,103 | 3,663 | -11% | 1 | 1 | 0% | 766 | 761 | -1% | 0 | 0 | — |
case-25 | fail→pass | 13,175 | 9,471 | -28% | 1 | 1 | 0% | 2,294 | 2,314 | +1% | 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 +76 percentage points is the difference between those two pass rates over the 25 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.