privacy-by-design – skill Thiết kế & UI cho agent AI

Privacy by Design

Overview

Integrate privacy protections into software architecture from the beginning, not as an afterthought. This skill applies Privacy by Design principles (GDPR Article 25, Cavoukian's framework) when designing databases, APIs, and user flows. Protects real users' data and builds trust.

When to Use This Skill

• Use when building apps that collect personal data (names, emails, locations, preferences)
• Use when designing database schemas, APIs, or authentication flows
• Use when the user mentions forms, user accounts, analytics, or third-party integrations
• Use when deploying to production—verify privacy controls before launch

Legal Frameworks

GDPR (EU) — Primary reference. Article 25 mandates "data protection by design and by default." Applies to EU users and often adopted globally.

CCPA (California) — Right to know, delete, opt-out of sale. Similar principles: minimize, disclose, allow control.

LGPD (Brazil) — Aligned with GDPR. Purpose limitation, necessity, transparency. Applies to Brazil users.

Map each applicable jurisdiction and processing purpose separately; compliance with one framework does not establish compliance with another.

---

Core Principles

1. Data Minimization

Collect only what is strictly necessary. Every field needs a documented justification. Avoid "we might need it later."

2. Purpose Limitation

Store the purpose of each data point. Identify and document the lawful basis and compatibility of every processing purpose; consent is one basis, not the only one.

3. Storage Limitation

Define retention periods. Implement automated deletion or anonymization when retention expires. Never keep data "forever" by default.

4. Privacy as Default

Opt-in for optional collection, not opt-out. Sensitive settings (analytics, marketing) off by default. No pre-checked consent boxes.

5. End-to-End Security

Encrypt at rest and in transit. Use RBAC. Log access to sensitive data for audit.

6. Transparency

Document what is collected and why. Clear privacy policies. Easy access and deletion for users.

---

User Rights (GDPR)

Ensure these are implementable from day one:

| Right | What to build |

|-------|---------------|

| Access | Endpoint or flow to return all user data |

| Rectification | Ability to update/correct data |

| Erasure | Verified erasure workflow, applicable legal exceptions and bounded backup expiry/restore suppression |

| Portability | Export data in machine-readable format (JSON, CSV) |

---

Deep Dive: Why It Matters

Data minimization — Less data = less breach impact, lower storage cost, simpler compliance. Each field is a liability.

Purpose limitation — Assess the lawful basis and compatibility of reuse under GDPR; consent is not universally required for every processing activity. Document purpose in schema or metadata.

Retention — Indefinite storage increases risk and violates GDPR. Define retention_days per data type; automate cleanup.

Logging — Logs often leak PII. Redact emails, IDs, tokens. Use structured logging with allowlists.

Third parties — Every SDK (analytics, crash reporting, ads) may send data elsewhere. Audit dependencies and establish the applicable consent/lawful-basis requirements before loading optional tracking.

---

Code Examples

JavaScript/Node — Minimal User Model

// BAD: Collecting everything "just in case"
const user = { email, name, phone, address, birthdate, ipAddress, userAgent, ... };

// GOOD: Minimal, documented purpose
const user = {
  email,        // purpose: authentication
  displayName,  // purpose: UI display
  createdAt,    // purpose: account age
};

JavaScript — Consent Before Tracking

// BAD: Track first, ask later
analytics.track(userId, event);

// GOOD: Check consent first
if (userConsent.analytics) {
  analytics.track(userId, event);
}

Python — Safe Logging

# BAD: Logging PII in plain text
logger.info(f"User {user.email} logged in from {request.remote_addr}")

# GOOD: Redact or hash identifiers
logger.info(f"User {hash_user_id(user.id)} logged in")
# Or: logger.info("User login", extra={"user_id_hash": hash_id(user.id)})

SQL — Schema with Purpose and Retention

-- GOOD: Document purpose and retention in schema
CREATE TABLE users (
  id UUID PRIMARY KEY,
  email VARCHAR(255) NOT NULL,  -- purpose: auth, retention: account lifetime
  display_name VARCHAR(100),   -- purpose: UI, retention: account lifetime
  created_at TIMESTAMPTZ,      -- purpose: audit, retention: 7 years
  last_login_at TIMESTAMPTZ    -- purpose: security, retention: 90 days
);

-- Add retention policy (PostgreSQL example)
-- Schedule job to anonymize/delete last_login_at after 90 days

API — Return Only Needed Fields

# BAD: Returning full user object
return jsonify(user)  # May include internal fields, hashed passwords

# GOOD: Explicit allowlist
return jsonify({
    "id": user.id,
    "email": user.email,
    "displayName": user.display_name,
})

---

Common Pitfalls

| Pitfall | Solution |

|---------|----------|

| Logs contain emails, IPs, tokens | Redact PII; use hashed IDs or structured logs |

| Error messages expose data | Return generic errors to client; log details server-side |

| Third-party SDKs load before consent | Load analytics/ads only after consent; use consent management |

| No deletion flow | Design account deletion + data purge from day one |

| Backups keep data forever | Include backups in retention; encrypt backups |

| Cookies without consent | Use consent banner; respect Do Not Track where applicable |

---

Third-Party Audit

Before adding a dependency that touches user data:

• [ ] What data does it collect or receive?
• [ ] Where does it send data (servers, countries)?
• [ ] Is it loaded before or after user consent?
• [ ] Can we disable it if user opts out?
• [ ] Does their privacy policy align with ours?

---

Implementation Checklist

When building a feature that touches user data:

• [ ] Is this data necessary? Can we achieve the goal with less?
• [ ] Do we have explicit consent for this use?
• [ ] Is it encrypted (at rest and in transit)?
• [ ] Do we have a retention/deletion policy?
• [ ] Can the user export or delete their data?
• [ ] Are third-party services disclosed and consented?
• [ ] Are logs free of PII?
• [ ] Are backups included in retention policy?

---

Best Practices

• ✅ Ask "do we need this?" for every new data field
• ✅ Design deletion and export flows from day one
• ✅ Use hashing or tokenization for sensitive identifiers when possible
• ✅ Document purpose and retention in schema or metadata
• ❌ Don't log passwords, tokens, or PII in plain text
• ❌ Don't share personal data without a documented lawful basis, appropriate safeguards and applicable consent
• ❌ Don't assume "we'll add privacy later"—it rarely happens
• ❌ Don't expose stack traces or internal errors to clients

---

When to Use

This skill is applicable when building software that collects, stores, or processes personal data. Apply it proactively during design and implementation.

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.

Hashing/tokenization usually pseudonymizes identifiers; it does not automatically anonymize data or remove privacy obligations. Assess re-identification risk and restrict keys/mappings. Retention and erasure need purpose-specific rules, lawful exceptions and tested restore/deletion handling. This is design guidance, not a compliance certification. See [EDPB lawful bases](https://www.edpb.europa.eu/topics/key-gdpr-concepts/legal-basis_en) and [ICO pseudonymisation](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-%72esources/data-sharing/anonymisation/pseudonymisation).