agentsclimarketplace

Gdpr us multinational

Skill vaquarkhan/compliance-agent-skills/skills/gdpr-us-multinational

30 Agent Skills for deterministic USA compliance auditing of AI agents — HIPAA, NIST AI RMF, FERPA, COPPA, PCI-DSS, SOC 2, FedRAMP, CMMC, GLBA, privacy & GDPR. Presidio PHI redaction, MCP templates, Pydantic AI, VS Code & JetBrains plugins.

Install
npx -y skills add vaquarkhan/compliance-agent-skills --skill gdpr-us-multinational

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Implements GDPR compliance workflows for US-headquartered multinationals—EU/EEA/UK data subjects, Articles 5-7 lawful basis, Article 30 records of processing, Article 32 security, Articles 33-34 breach notification (72 hours), Article 35 DPIA, Standard Contractual Clauses, EU-US Data Privacy Framework adequacy, and DPA/subprocessor governance for agent/LLM cross-border transfers. Trigger when US companies process EU residents' data, deploying agents in EU regions, assessing LLM vendor international transfers, or preparing RoPA/DPIA for multinational privacy programs. Do not use for US state laws only (use us-state-privacy-laws or ccpa-cpra-privacy-rights), HIPAA PHI (use hipaa-technical-safeguards), or UK-only post-Brexit without EU scope (note UK GDPR parallels in references).

SKILL.md

9.8 KB, as published. Nobody here has run it

GDPR for US Multinationals

Overview

General Data Protection Regulation (EU) 2016/679 applies to US companies when they offer goods/services to EU/EEA data subjects or monitor their behavior—regardless of US headquarters. Agent/LLM systems processing EU user prompts, support chat, or analytics fall in scope.

GDPR topicArticleUS multinational focus
PrinciplesArt. 5Lawfulness, minimization, storage limitation for agent logs
Lawful basisArt. 6Consent vs legitimate interest for AI features
Special categoriesArt. 9Health, biometric in prompts—explicit consent or Art. 9(2) exception
Records of processingArt. 30RoPA must list LLM/MCP subprocessors and transfers
SecurityArt. 32Encryption, redaction gate, MCP access controls
Breach notificationArt. 33-3472 hours to supervisory authority; data subject notice if high risk
DPIAArt. 35Required for systematic profiling, large-scale special categories
TransfersArt. 44-49SCCs, DPF, or adequacy for US LLM processing
DPOArt. 37-39Required for large-scale special category or systematic monitoring
RightsArt. 15-22Access, erasure, portability—LLM vendor deletion complexity

Transfer mechanisms (2024+):

  • EU-US Data Privacy Framework (DPF) — US entities self-certify; verify LLM vendor DPF status
  • Standard Contractual Clauses (2021 SCCs) — Module 2 controller-processor for LLM vendors
  • UK IDTA / Addendum — if UK data subjects included

Reference: references/gdpr-article-checklist.md, template templates/gdpr-ropa-template.yaml

When to Use

Use this skill when:

  • US company processes EU/EEA (or UK) residents' personal data in products or agents
  • Assessing cross-border transfers to US-based LLM APIs (OpenAI, Anthropic, etc.)
  • Building Article 30 RoPA including agent/MCP processing activities
  • DPIA for agent profiling, automated decisions, or health-related prompts
  • 72-hour breach notification to lead supervisory authority (cross-border one-stop-shop)
  • Drafting DPA/SCCs with LLM and MCP subprocessors
  • DPO consultation on high-risk AI agent deployment in EU market

Do not use this skill when:

  • US-only processing (use us-state-privacy-laws, ccpa-cpra-privacy-rights)
  • HIPAA-covered PHI for US treatment/payment (HIPAA may apply; GDPR may also apply for EU patients—dual track)
  • FedRAMP/CMMC without EU data (use respective skills)

Core Process

Execute steps in order.

Step 1: Territorial scope and roles

  1. Confirm Art. 3 applicability: EU establishment, offering goods/services to EU, or monitoring behavior.
  2. Identify controller, joint controller, processor roles:
    • US SaaS with EU customers → typically controller
    • LLM API → processor (or subprocessor)
  3. Determine lead supervisory authority (Art. 56) if main establishment in EU; US-only establishment → identify EU member state of most data subjects or first contact.
  4. Artifact: scope JSON with establishment map.

Step 2: Lawful basis (Art. 6)

For each agent processing activity, document lawful basis:

ActivityTypical basisNotes
Customer support agentContract Art. 6(1)(b)Necessary for service
Product analytics on promptsLegitimate interest Art. 6(1)(f)LIA required; opt-out where applicable
Marketing personalization agentConsent Art. 6(1)(a)Granular, withdrawable
Compliance audit agentLegal obligation Art. 6(1)(c)Document statute
  1. Special category data (Art. 9) in prompts—health, biometric: explicit consent or documented Art. 9(2) exception; default deny without basis.
  2. Apply redaction.py before US LLM transfer as technical minimization (Art. 5(1)(c)).

Step 3: Records of processing (Art. 30)

Populate templates/gdpr-ropa-template.yaml:

  1. Processing activity name (e.g., "EU Customer Support Agent")
  2. Purposes, categories of data subjects and personal data
  3. Recipients including US LLM subprocessors
  4. Transfers to third countries + safeguards (SCCs, DPF)
  5. Retention periods; security measures (Art. 32 cross-ref)
  6. Do not store raw special category data in RoPA—reference categories only.

Step 4: International transfers (Chapter V)

  1. Inventory transfers to US for LLM inference, logging, training:
    • Vendor DPF certification status (DataPrivacyFramework.gov)
    • If no DPF: execute 2021 SCCs Module 2 (controller-processor) + UK Addendum if needed
  2. Transfer impact assessment (TIA) for US surveillance laws post-Schrems II:
    • Supplementary measures: encryption in transit, redaction, contractual audit, data residency EU region if offered
  3. MCP servers in US: document as subprocessor transfers.

Step 5: Data subject rights (Arts. 15-22)

RightAgent/LLM implementation
Access Art. 15Export prompts/logs; redact third-party PI
Erasure Art. 17Delete from DB, vector store; vendor deletion SLA
Portability Art. 20JSON export of agent session data
Object Art. 21Stop profiling-based agent features
Restrict Art. 18Flag account—no LLM processing pending review
  1. One-month response (extendable 2 months)—stricter than many US states.
  2. Route DSARs without sending full request to US LLM unredacted.

Step 6: Security (Art. 32)

  1. Align with hipaa-phi-redaction-pipeline and access-control-identity-audit where applicable.
  2. Document: pseudonymization (Presidio tokens), encryption, MCP OAuth, resilience testing.
  3. Deanonymization opt-in in agent.py supports Art. 32 minimization—default redacted output.

Step 7: DPIA (Art. 35)

Required when processing likely high risk—systematic profiling, large-scale special categories, public monitoring:

  1. Describe agent processing, necessity, proportionality
  2. Risk to data subjects: incorrect automated decisions, US transfer exposure
  3. Mitigations: human review, redaction, EU region deployment, SCCs
  4. Consult DPO (Art. 35(2)); seek supervisory authority opinion if residual high risk (Art. 36)

Step 8: Breach notification (Arts. 33-34)

  1. 72 hours from awareness to supervisory authority—not 60 days like HIPAA.
  2. If high risk to individuals: communicate breach to data subjects without undue delay (Art. 34).
  3. Cross-border: notify lead authority; cooperate with concerned authorities (Art. 33(2)).
  4. Use breach-incident-response for containment; this skill for GDPR clock and content.
  5. Document processor notification from LLM vendor (Art. 33(2) processor→controller without undue delay).

Step 9: Processor agreements (Art. 28)

LLM/MCP DPA must include:

  • Subject matter, duration, nature/purpose, data types, controller instructions
  • Confidentiality, security Art. 32, subprocessor authorization
  • Assistance with DSAR, DPIA, breach notification
  • Deletion/return at end of service
  • Audit and inspection rights

Step 10: Documentation and accountability (Art. 5(2))

  1. Maintain RoPA, DPIA, LIA, SCCs, DPF certifications, training records.
  2. Evidence manifest with SHA-256 for audit readiness.
  3. Disclaim: operational readiness—not legal opinion; engage EU privacy counsel for binding interpretation.

Common Rationalizations

ExcuseRebuttal
"We're US-only company—GDPR doesn't apply."Art. 3(2) applies if offering services to EU data subjects—SaaS with EU users triggers GDPR.
"LLM vendor handles GDPR—we're processor only."SaaS owner is usually controller for customer relationship—cannot outsource accountability.
"HIPAA compliance satisfies GDPR."GDPR has broader rights, 72-hour breach, DPIA, and transfer rules—separate program required.
"Redaction means no transfer of personal data."If re-identification possible or tokens map stored, still personal data—document transfer mechanism.
"SCCs alone are enough post-Schrems II."TIA + supplementary measures required—document encryption and access controls.

Red Flags

  • EU user prompts sent to US LLM without SCCs/DPF/TIA
  • No RoPA entry for agent processing activities
  • Art. 9 special category data in prompts without documented basis
  • Breach response plan uses HIPAA 60-day timeline only
  • No DPO involvement in agent DPIA despite profiling
  • Deanonymization enabled by default in production EU path (violates minimization)

Verification

  • Art. 3 applicability and controller/processor roles documented
  • Lawful basis (and Art. 9 if applicable) for each agent activity
  • Art. 30 RoPA populated including US transfers and subprocessors
  • Transfer mechanism verified (DPF/SCCs) with TIA on file
  • DSAR workflow meets one-month deadline with LLM vendor chain
  • Art. 32 security measures mapped to technical controls
  • DPIA completed for high-risk agent processing; DPO consulted
  • 72-hour breach notification playbook integrated
  • Art. 28 processor DPAs executed with LLM/MCP vendors

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.