Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When the user needs to draft, review, or update a privacy policy for their product, or needs to understand data privacy obligations across jurisdictions.
.claude/skills/thomasmoreai-privacy-policy-mkurman/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 11 |
| gemini-3.1-pro-preview | 100% | 4 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 79% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 97% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 82% | 0% |
Activate when a founder needs to create a privacy policy for a new product launch, update an existing policy for new data practices or features, expand into a new jurisdiction (EU, California, etc.), or assess whether current data handling is properly disclosed. Also activate when the user asks about GDPR, CCPA, CPRA, or general data privacy compliance.
[LEGAL REVIEW REQUIRED] notation. These include legal basis determinations, international transfer mechanisms, and jurisdiction-specific rights.Three-part deliverable:
Product details, data types collected, applicable jurisdictions, user rights summary, retention overview, and contact information.
terms-of-service -- Draft alongside the privacy policy; they should cross-reference each other and use consistent definitions.soc2-prep -- SOC 2 Trust Service Criteria for Privacy directly overlaps with privacy policy commitments.security-review -- Security measures described in the privacy policy must reflect actual technical controls.User: "We're launching our project management SaaS next month with users from the US and Europe. We use Stripe, Mixpanel, and AWS."
Good output: A three-part deliverable. The data inventory table mapping each data category to collection method, purpose, legal basis, third parties, and retention. Jurisdiction analysis identifying GDPR applicability and CCPA threshold monitoring. Red flags for Mixpanel IP collection needing disclosure and DPA, and missing cookie consent mechanism for EU users.
User: "We added an AI assistant that processes customer messages. Do we need to update our privacy policy?"
Good output: Identifies the new data processing (message content processed by AI models), new third party (AI provider as sub-processor), new legal basis analysis needed, and GDPR Art. 22 consideration for automated decision-making. Provides the specific policy sections that need updating with draft language.
Disclaimer: This skill generates draft privacy policies and compliance guidance for educational and planning purposes only. It does not constitute legal advice. Always have a qualified attorney licensed in your relevant jurisdictions review the final privacy policy before publication. Regulatory non-compliance can result in significant fines (up to 4% of global annual revenue under GDPR).
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 29,357 | 27,388 | -7% | 1 | 1 | 0% | 6,260 | 7,220 | +15% | 0 | 0 | — |
case-01 | fail→pass | 25,536 | 27,070 | +6% | 1 | 1 | 0% | 4,930 | 7,716 | +57% | 0 | 0 | — |
case-03 | fail→fail | 30,391 | 27,600 | -9% | 1 | 1 | 0% | 6,263 | 7,559 | +21% | 0 | 0 | — |
case-04 | pass→pass | 17,398 | 24,529 | +41% | 1 | 1 | 0% | 3,829 | 6,999 | +83% | 0 | 0 | — |
case-05 | pass→pass | 24,047 | 23,039 | -4% | 1 | 1 | 0% | 4,555 | 6,163 | +35% | 0 | 0 | — |
case-06 | pass→pass | 19,360 | 26,504 | +37% | 1 | 1 | 0% | 3,605 | 6,376 | +77% | 0 | 0 | — |
case-07 | pass→pass | 30,198 | 31,526 | +4% | 1 | 1 | 0% | 5,501 | 7,520 | +37% | 0 | 0 | — |
case-08 | fail→pass | 21,744 | 29,384 | +35% | 1 | 1 | 0% | 3,800 | 6,787 | +79% | 0 | 0 | — |
case-09 | pass→pass | 25,205 | 31,116 | +23% | 1 | 1 | 0% | 4,841 | 7,485 | +55% | 0 | 0 | — |
case-10 | pass→pass | 24,141 | 24,335 | +1% | 1 | 1 | 0% | 4,459 | 6,499 | +46% | 0 | 0 | — |
case-11 | fail→fail | 21,819 | 33,095 | +52% | 1 | 1 | 0% | 4,245 | 7,735 | +82% | 0 | 0 | — |
case-12 | fail→pass | 17,368 | 23,777 | +37% | 1 | 1 | 0% | 3,455 | 6,819 | +97% | 0 | 0 | — |
case-13 | pass→pass | 18,719 | 19,768 | +6% | 1 | 1 | 0% | 3,387 | 5,471 | +62% | 0 | 0 | — |
case-14 | fail→pass | 17,612 | 20,073 | +14% | 1 | 1 | 0% | 3,065 | 5,566 | +82% | 0 | 0 | — |
case-15 | fail→pass | 33,126 | 27,528 | -17% | 1 | 1 | 0% | 6,186 | 6,912 | +12% | 0 | 0 | — |
case-16 | fail→pass | 15,922 | 25,738 | +62% | 1 | 1 | 0% | 2,811 | 6,616 | +135% | 0 | 0 | — |
case-17 | fail→fail | 14,939 | 22,547 | +51% | 1 | 1 | 0% | 2,635 | 5,860 | +122% | 0 | 0 | — |
case-18 | fail→pass | 32,622 | 24,552 | -25% | 1 | 1 | 0% | 6,182 | 6,673 | +8% | 0 | 0 | — |
case-19 | fail→fail | 11,273 | 26,080 | +131% | 1 | 1 | 0% | 2,429 | 7,026 | +189% | 0 | 0 | — |
case-20 | pass→pass | 21,356 | 23,429 | +10% | 1 | 1 | 0% | 4,022 | 6,294 | +56% | 0 | 0 | — |
case-21 | pass→pass | 18,610 | 32,032 | +72% | 1 | 1 | 0% | 3,493 | 7,937 | +127% | 0 | 0 | — |
case-22 | fail→pass | 12,925 | 25,363 | +96% | 1 | 1 | 0% | 2,758 | 6,762 | +145% | 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 +41 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.