Privacy review

privacy-review

When to use

• A new feature, integration, or vendor introduces a data flow and the question is which regulatory regime applies, what consent / lawful basis is required, what the breach-impact tail looks like.
• An existing data flow is being re-scoped (new geography, new customer segment, new processor) and the regulatory-regime delta must be read before the change ships.
• A customer / counterparty requests a DPA, BAA, or SCC and the question is what we can credibly sign given current data-handling reality.

Do NOT use as a substitute for qualified privacy counsel (this skill produces the non-lawyer cognition that prepares the counsel conversation), as a contract-level read (route to contracts-cognition (P5); P5 composes this skill for data-clause depth), or for privacy-platform SaaS configuration / audit-tool administration.

Cognition cluster

• Mental model 28 — Inversion. "What's the worst-case if this data flow leaks, is subpoenaed, or is mis-consented?" Inversion sizes the regulatory tail before consent / DPA shape is debated. See mental-models.md § 28.
• Mental model 1 — First principles. Strip the data flow to: who (data subject), what (data category), why (lawful basis / purpose), where (residency / transfer), how long (retention), who else (sub-processors). Six primitives anchor every regime delta. See mental-models.md § 1.
• Mental model 21 — Second-order thinking. Consent design at signup interacts with marketing automation; retention defaults interact with data-subject-rights workflow; sub-processor chains interact with breach-notification timelines. Each privacy choice has downstream regime obligations. See mental-models.md § 21.
• Context-spine — regulatory-regime + customer-segment + product. Read regulatory-regime (J1) for the applicable floor (GDPR, CCPA/CPRA, HIPAA, PIPEDA, LGPD, sector-specific). Read customer-segment for who the data subjects are (B2C-EU = GDPR primary; US-healthcare = HIPAA primary; B2B-EU-of-US-customers = mixed). Read product for which features touch sensitive categories.

Cross-wing handoff

• Cites J1 regulatory-regime (foundation slot) for the regime floor read; without J1, this skill cannot bind which regime applies.
• Hands off to P5 contracts-cognition for clause-level redlines on DPAs / BAAs / SCCs.
• Hands off to P7 data-handling-judgment for the classification, retention, and cross-border-transfer surface that this skill flags.

Procedure

Step 0: Bind the regulatory-regime floor

Read the J1 regulatory-regime slot for the customer-segment and geography in scope. For each applicable regime, name:

1. Which data subjects does it protect? (EU residents → GDPR; California residents → CCPA/CPRA; US patients in covered entities → HIPAA).
2. What categories does it gate? (GDPR special categories; HIPAA PHI; CCPA "sensitive personal information").
3. What lawful-basis / consent shape does it require? (GDPR Art. 6 + Art. 9; HIPAA authorization; CCPA notice + opt-out).

Multiple regimes can apply simultaneously; the floor is the strictest applicable.

Step 1: Map the data flow

For the feature / integration / vendor in scope, enumerate every hop:

1. Collection — what data category, from whom, where, with what notice / consent.
2. Processing — who processes (us, sub-processor), where, for what purpose.
3. Storage — where stored (region, system), encrypted at rest, retention default.
4. Transfer — cross-border hops, transfer mechanism (SCC, adequacy, BCR).
5. Disclosure — who sees it (internal roles, third parties, government-access risk).
6. Deletion — retention end-state, data-subject-rights workflow, hard-delete vs soft-delete.

A flow without all six hops named is not mapped; it's assumed. Force the enumeration.

Step 2: Compute the regulatory-regime delta

For each hop × each applicable regime, name the obligation:

1. Lawful basis / consent — what's the basis for this hop under this regime; is it documented; is it user-affirmative where required.
2. Notice — what notice was given at the collection hop; does it cover the downstream hops.
3. DPA / BAA / SCC — for each sub-processor / cross-border hop, what contract is required; is it in place.
4. Data-subject rights — for each hop, can we deliver access / deletion / portability / objection within the regime's window.
5. Breach-notification surface — what's the notification timeline (GDPR 72 h, HIPAA 60 d, CCPA varies), to whom, with what content.

Gaps = obligations un-met. Surface them, don't smooth them.

Step 3: Consent design read

For each lawful-basis claim:

1. Is the basis defensible? (legitimate-interest balancing test documented; consent freely given / specific / informed / unambiguous; contractual necessity actually necessary).
2. Is withdrawal as easy as granting? (GDPR Art. 7.3; CCPA opt-out parity).
3. Are sensitive categories handled with explicit consent / authorization?

Consent-as-checkbox-at-signup is the canonical failure mode. Inversion check: "if a regulator inspects the consent flow tomorrow, what would they find un-defensible?"

Step 4: Breach-impact triage

For each data category × hop, size the breach tail:

1. Volume — how many subjects per hop.
2. Sensitivity — special-category / PHI / financial / identity-document presence.
3. Notification surface — 72-hour clock starts when; who notifies (us as controller, vendor as processor); content requirements.
4. Regulatory-fine exposure — GDPR up to 4 % global turnover or €20M; HIPAA tiered; CCPA per-record statutory damages.

A hop without a sized breach-tail is unstressed.

Step 5: Validate the privacy read before emitting

Before producing the artifact, verify three things:

1. Hop coverage — confirm all six data-flow hops (collection, processing, storage, transfer, disclosure, deletion) were inspected; silent skips mean the flow was not mapped.
2. Regime delta completeness — assert every applicable regime was checked for lawful basis, notice, DPA/BAA/SCC, data-subject rights, breach notification; un-checked obligations are gaps in disguise.
3. Counsel handoff — verify the artifact explicitly flags which findings need privacy-counsel sign-off vs which are operational decisions; this skill does not replace counsel.

All three must pass. If any fails, return to the failing step.

Step 6: Emit the privacy-review note

Produce the privacy-review artifact for the feature owner, legal / counsel, and DPO if applicable. The artifact is the non-lawyer cognition that prepares the counsel conversation and gates the ship decision; it is not the legal opinion.

Related Skills

WHEN to use this

• Reviewing a new feature / integration / vendor data flow against applicable regulatory regimes.
• Re-scoping an existing flow under a new geography, segment, or processor.
• Sizing the breach-impact tail before a ship decision.

WHEN NOT to use this

• Contract-clause redline depth — route to contracts-cognition (P5); P5 composes this skill for data-clause sections.
• Data-classification / retention / cross-border judgment in isolation — route to data-handling-judgment (P7); P7 is composed by this skill.
• Regulatory-regime floor read in general — route to regulatory-regime (J1); this skill cites J1, doesn't replace it.
• Legal privacy opinion — route to qualified privacy counsel.

When the agent should load this

• "Is this GDPR-safe?"
• "Do we need a DPA / BAA / SCC for this vendor?"
• "What's the breach exposure on this data flow?"
• "Review the consent flow for the new signup."
• "Wir starten in der EU — was ändert sich datenschutzrechtlich?"

Output

1. data-flow-map.md — six hops × what / who / where / how-long / who-else per hop.
2. regime-delta.md — applicable regimes × obligations × gaps per hop.
3. consent-design-note.md — lawful-basis defensibility, withdrawal parity, sensitive-category handling.
4. breach-impact-triage.md — sized tail per category × hop; notification clock; fine exposure.
5. counsel-handoff.md — findings that need counsel sign-off vs operational decisions.

Gotcha

• "We're US-only, GDPR doesn't apply" is often wrong; GDPR follows the data subject, not the company.
• Consent checkboxes pre-ticked at signup are not consent under GDPR; this is a canonical regulator-attention pattern.
• Sub-processor chains drift silently; a vendor adds a sub-processor 6 months in and your DPA is stale.
• Retention defaults of "forever" interact badly with data-subject-rights timelines under every regime.

Do NOT

• Do NOT issue privacy legal opinions; this skill prepares cognition for counsel, not replaces counsel.
• Do NOT collapse multiple regimes into the most familiar one; the floor is the strictest applicable.
• Do NOT skip breach-impact triage; un-stressed flows ship with un-sized tail.

Runnable example

Growth-stage SaaS adds an EU customer cohort; existing US-only flow now serves EU data subjects.

• Step 0 — Bind regime: GDPR applies (EU data subjects); CCPA still applies for California subset; HIPAA n/a (no PHI).
• Step 1 — Map flow: collection (signup, EU IP), processing (US-east region), storage (US-east + analytics warehouse), transfer (US-east → analytics vendor in US-west; analytics vendor uses sub-processor in India), disclosure (internal CS team, no third parties), deletion (soft-delete, 7-year retention default).
• Step 2 — Regime delta: lawful basis missing for analytics (no opt-in); notice doesn't disclose Indian sub-processor; no SCC with analytics vendor; data-subject-rights workflow is manual ticket only; breach-notification process undocumented.
• Step 3 — Consent: signup checkbox is pre-ticked for marketing — fails Art. 7. Withdrawal requires support email — fails parity.
• Step 4 — Breach-impact: 12k EU subjects, no special categories, GDPR fine exposure up to 4 % global turnover; 72-hour notification clock with no documented owner.
• Step 5 — Validate: six hops mapped; both regimes checked; counsel-handoff names SCC + analytics-vendor sub-processor chain + consent UX redesign as counsel-led; retention default + DSR workflow as operational. Pass.
• Step 6 — Emit privacy-review note; gate EU launch on (a) SCC with analytics vendor; (b) consent UX redesign; (c) DSR workflow with named owner and 30-day window; (d) breach-notification runbook.