Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Navigate privacy regulations (GDPR, CCPA), review DPAs, and handle data subject requests. Use when reviewing data processing agreements, responding to data subject access or deletion requests, assessing cross-border data transfer requirements, or evaluating privacy compliance.
.claude/skills/vythanhtra-legal-compliance/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | -18% | 0% |
| case-08 | ✓→✓ | ＝ Same ✓ | -6% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -37% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 73% | 0% |
You are a compliance assistant for an in-house legal team. You help with privacy regulation compliance, DPA reviews, data subject request handling, and regulatory monitoring.
Scope: Applies to processing of personal data of individuals in the EU/EEA, regardless of where the processing organization is located.
Key Obligations:
Key Obligations:
| Regulation | Initial Acknowledgment | Substantive Response | Extension | |---|---|---|---| | GDPR | Promptly (best practice) | 30 days | +60 days | | CCPA/CPRA | 10 business days | 45 calendar days | +45 days | | UK GDPR | Promptly (best practice) | 30 days | +60 days |
What to Monitor:
Escalation Criteria:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 15,968 | 9,992 | -37% | 1 | 1 | 0% | 2,676 | 2,201 | -18% | 0 | 0 | — |
case-02 | fail→fail | 17,711 | 15,125 | -15% | 1 | 1 | 0% | 3,076 | 3,365 | +9% | 0 | 0 | — |
case-08 | pass→pass | 10,800 | 6,538 | -39% | 1 | 1 | 0% | 2,098 | 1,973 | -6% | 0 | 0 | — |
case-03 | fail→fail | 15,188 | 15,104 | -1% | 1 | 1 | 0% | 2,656 | 3,521 | +33% | 0 | 0 | — |
case-04 | pass→pass | 18,458 | 7,134 | -61% | 1 | 1 | 0% | 3,071 | 1,922 | -37% | 0 | 0 | — |
case-05 | pass→pass | 5,452 | 5,652 | +4% | 1 | 1 | 0% | 1,031 | 1,784 | +73% | 0 | 0 | — |
case-06 | pass→pass | 6,901 | 5,185 | -25% | 1 | 1 | 0% | 1,320 | 1,823 | +38% | 0 | 0 | — |
case-07 | pass→pass | 11,690 | 6,179 | -47% | 1 | 1 | 0% | 2,000 | 1,880 | -6% | 0 | 0 | — |
case-09 | fail→pass | 8,383 | 7,162 | -15% | 1 | 1 | 0% | 1,476 | 1,999 | +35% | 0 | 0 | — |
case-10 | pass→pass | 5,721 | 5,701 | -0% | 1 | 1 | 0% | 1,015 | 1,778 | +75% | 0 | 0 | — |
case-11 | pass→pass | 11,296 | 7,682 | -32% | 1 | 1 | 0% | 2,016 | 2,122 | +5% | 0 | 0 | — |
case-12 | pass→pass | 11,247 | 8,333 | -26% | 1 | 1 | 0% | 1,887 | 2,280 | +21% | 0 | 0 | — |
case-13 | pass→pass | 7,027 | 5,635 | -20% | 1 | 1 | 0% | 1,255 | 1,807 | +44% | 0 | 0 | — |
case-14 | pass→pass | 13,166 | 6,256 | -52% | 1 | 1 | 0% | 2,491 | 1,920 | -23% | 0 | 0 | — |
case-15 | pass→pass | 14,195 | 7,594 | -47% | 1 | 1 | 0% | 2,411 | 2,097 | -13% | 0 | 0 | — |
case-16 | pass→pass | 11,089 | 9,063 | -18% | 1 | 1 | 0% | 1,972 | 2,480 | +26% | 0 | 0 | — |
case-17 | pass→pass | 6,872 | 7,694 | +12% | 1 | 1 | 0% | 1,202 | 2,154 | +79% | 0 | 0 | — |
case-18 | pass→pass | 13,579 | 6,750 | -50% | 1 | 1 | 0% | 2,107 | 1,840 | -13% | 0 | 0 | — |
case-19 | pass→pass | 12,659 | 7,559 | -40% | 1 | 1 | 0% | 1,623 | 1,982 | +22% | 0 | 0 | — |
case-20 | pass→pass | 10,078 | 7,476 | -26% | 1 | 1 | 0% | 1,870 | 2,204 | +18% | 0 | 0 | — |
case-21 | pass→pass | 7,831 | 5,310 | -32% | 1 | 1 | 0% | 1,425 | 1,727 | +21% | 0 | 0 | — |
case-22 | pass→pass | 11,353 | 12,580 | +11% | 1 | 1 | 0% | 1,828 | 2,658 | +45% | 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.