Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Australian IRAP (Information Security Registered Assessors Program) expert. Provides guidance on ISM controls, Essential Eight maturity levels, ACSC guidelines, and Australian data sovereignty requirements.
.claude/skills/grcengclub-irap-expert/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-11 | ✓→✓ | ＝ Same ✓ | -4% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 1% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 174% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 53% | 0% |
Expertise in Australian government cloud security based on ISM and Essential Eight.
Authority: Australian Cyber Security Centre (ACSC) Base Standard: Information Security Manual (ISM) Key Framework: Essential Eight maturity model
Scope: Australian government agencies and contractors
| Level | Use Case | Residency | |-------|----------|-----------| | OFFICIAL | Routine business | No requirement | | OFFICIAL:Sensitive | Personal info | Recommended AU | | PROTECTED | Cabinet, national security | AU regions mandatory | | SECRET | Intelligence | AU regions mandatory | | TOP SECRET | Highest sensitivity | Dedicated infrastructure |
Maturity Levels:
Over 1,400 security controls organized by:
Region: ap-southeast-2 (Sydney) Requirement: PROTECTED data must stay in Australia
Process:
Assessors: ACSC-endorsed IRAP assessors
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | pass→pass | 13,540 | 10,374 | -23% | 1 | 1 | 0% | 2,055 | 1,968 | -4% | 0 | 0 | — |
case-01 | pass→pass | 16,558 | 18,646 | +13% | 1 | 1 | 0% | 2,473 | 2,491 | +1% | 0 | 0 | — |
case-02 | fail→fail | 9,242 | 8,595 | -7% | 1 | 1 | 0% | 1,503 | 1,859 | +24% | 0 | 0 | — |
case-03 | pass→pass | 3,753 | 6,159 | +64% | 1 | 1 | 0% | 548 | 1,504 | +174% | 0 | 0 | — |
case-04 | pass→pass | 8,405 | 9,100 | +8% | 1 | 1 | 0% | 1,243 | 1,905 | +53% | 0 | 0 | — |
case-05 | pass→pass | 7,697 | 6,763 | -12% | 1 | 1 | 0% | 1,135 | 1,489 | +31% | 0 | 0 | — |
case-06 | pass→pass | 5,357 | 6,245 | +17% | 1 | 1 | 0% | 781 | 1,376 | +76% | 0 | 0 | — |
case-07 | pass→pass | 11,176 | 10,384 | -7% | 1 | 1 | 0% | 1,675 | 2,118 | +26% | 0 | 0 | — |
case-08 | pass→pass | 8,781 | 9,664 | +10% | 1 | 1 | 0% | 1,318 | 1,996 | +51% | 0 | 0 | — |
case-09 | pass→pass | 15,261 | 12,950 | -15% | 1 | 1 | 0% | 2,209 | 2,456 | +11% | 0 | 0 | — |
case-10 | pass→pass | 13,063 | 10,217 | -22% | 1 | 1 | 0% | 2,056 | 2,064 | +0% | 0 | 0 | — |
case-12 | pass→pass | 15,523 | 10,301 | -34% | 1 | 1 | 0% | 2,270 | 2,107 | -7% | 0 | 0 | — |
case-13 | pass→pass | 10,762 | 8,132 | -24% | 1 | 1 | 0% | 1,617 | 1,712 | +6% | 0 | 0 | — |
case-14 | pass→pass | 4,929 | 4,546 | -8% | 1 | 1 | 0% | 811 | 1,206 | +49% | 0 | 0 | — |
case-15 | fail→pass | 14,196 | 10,388 | -27% | 1 | 1 | 0% | 2,380 | 2,139 | -10% | 0 | 0 | — |
case-16 | pass→pass | 16,199 | 11,891 | -27% | 1 | 1 | 0% | 2,279 | 2,191 | -4% | 0 | 0 | — |
case-17 | pass→pass | 15,534 | 10,073 | -35% | 1 | 1 | 0% | 2,259 | 1,971 | -13% | 0 | 0 | — |
case-18 | pass→pass | 14,653 | 12,658 | -14% | 1 | 1 | 0% | 2,197 | 2,310 | +5% | 0 | 0 | — |
case-19 | pass→pass | 8,222 | 10,251 | +25% | 1 | 1 | 0% | 1,291 | 2,077 | +61% | 0 | 0 | — |
case-20 | pass→pass | 10,679 | 13,251 | +24% | 1 | 1 | 0% | 1,479 | 2,388 | +61% | 0 | 0 | — |
case-21 | pass→pass | 7,031 | 8,245 | +17% | 1 | 1 | 0% | 1,101 | 1,739 | +58% | 0 | 0 | — |
case-22 | pass→pass | 12,797 | 3,368 | -74% | 1 | 1 | 0% | 2,014 | 1,033 | -49% | 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 +5 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.