compliance-check

name	compliance-check
description	GDPR + ePrivacy audit — consent, privacy policy, data handling, right-to-deletion, DPAs. Deep-dive for /launch-check compliance.
disable-model-invocation	false
argument-hint	[project-path]
effort	high

/compliance-check — GDPR + ePrivacy Compliance

Deep-dive regulatory compliance analysis. Checks cookie consent, privacy policy, terms of service, data handling, and user rights. Invoke when /launch-check's compliance row shows WARN or FAIL, or proactively before launching in the EU/UK.

Consumes the DFD produced by /dfd for data-handling analysis (Step 4). The DFD's classifications table at projects/<project>/architecture/dfd.md is the source of truth for which fields are PII / PCI / secrets, and the DFD's external-services list is the source of truth for cross-border transfers and third-party processors. If the DFD doesn't exist, this skill falls back to its own grep-based discovery (degraded mode — surface in the output banner). See AgDR-0026 for the single-source-of-truth rationale.

Compliance Areas

Area	Regulation	Key requirement
Cookie consent	ePrivacy Directive	Informed consent before setting non-essential cookies
Privacy policy	GDPR Art. 13-14	Clear disclosure of what data is collected, why, and how long
Right to deletion	GDPR Art. 17	Users can request their data be deleted
Data minimization	GDPR Art. 5	Collect only what's necessary
Data retention	GDPR Art. 5	Don't keep data longer than needed
Terms of service	Contract law	Clear terms governing use of the service
Third-party data sharing	GDPR Art. 28	Data processing agreements with third parties

Process

Step 1: Cookie and tracking audit

• Grep for cookie-setting code (document.cookie, setCookie, cookie-parser, js-cookie)
• Grep for analytics SDKs (Google Analytics, Mixpanel, Amplitude, PostHog, Segment, Facebook Pixel)
• Grep for cookie consent libraries (cookieconsent, react-cookie-consent, cookie-banner, consent-manager)
• Check if consent is obtained BEFORE analytics/tracking scripts load (not after)
• Check for a cookie policy page listing all cookies and their purposes

Step 2: Privacy and legal pages

• Check for /privacy or /privacy-policy route
• Check for /terms or /terms-of-service route
• Check if privacy policy covers: what data is collected, legal basis, retention period, third parties, user rights, contact info
• Check for a link to the privacy policy in the sign-up flow and footer

Step 3: User rights implementation

• Check for a "delete account" or "delete my data" feature
• Check for a "download my data" / data portability feature
• Check for opt-out mechanisms for marketing communications
• Check if the auth system supports account deactivation

Step 4: Data handling (DFD-driven)

Prefer the DFD when present. Read the classifications table at projects/<project>/architecture/dfd.md § "Data classifications" — that's the canonical view of PII / PCI / secrets across the system, produced by /dfd's three-pathway heuristic walk (annotations + env-var heuristics + schema columns) plus any explicit registry the project ships.

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-read-config.sh"
source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-portfolio-paths.sh"
projects_dir=$(portfolio_projects_dir)

dfd="${projects_dir}/${project_name}/architecture/dfd.md"
if [ -f "$dfd" ]; then
  # DFD present — use it as source of truth
  : # parse the Classifications table + External actors block
else
  # Degraded mode — surface in output banner, fall back to grep walk
  : # legacy independent scan
fi

From the DFD's classifications + flows, derive:

• What PII is stored — every row labelled pii in the classifications table
• What PCI data crosses the system — every row labelled pci (any presence triggers stricter retention + encryption rules)
• Where secrets live — every row labelled secrets (env vars + schema columns) — confirm they're loaded from a secrets manager, not committed
• Cross-border transfers — every External Service actor + its inferred region (Stripe-US, SendGrid-US, Auth0-US, Cognito-region-dependent, etc.) — flag if any PII flow lands on a non-EU vendor without a Standard Contractual Clause
• Third-party processors — every External Service is a candidate Data Processing Agreement (DPA) requirement

Then complement with the inline checks (still needed regardless of DFD source):

• Check if PII is encrypted at rest (DB encryption flag / column-level encryption)
• Check if sensitive data is logged (grep for logging calls that might include user data — the DFD lists the fields; this step verifies they're not in logs)
• Check for data retention policies (is old data cleaned up? — usually only in cron specs or DB triggers)

Step 5: Output findings

COMPLIANCE CHECK — <project> @ <sha>

| # | Area | Status | Finding | Action |
|----|------|--------|---------|--------|
| C1 | Cookies | FAIL | Analytics loads before consent | Gate GA4 behind consent callback |
| C2 | Privacy | WARN | Policy exists but doesn't list retention periods | Add retention section |
| C3 | Deletion | FAIL | No delete-account endpoint | Add DELETE /api/user/me endpoint |
| C4 | Data minimization | PASS | Only email + name collected at signup | — |

Summary: <N> findings (<N> fail, <N> warn, <N> pass)
GDPR readiness: <READY / PARTIAL / NOT READY>

Persist the run + render trend

After printing the findings table (Step 5), persist a structured artefact via the shared audit-history lib so the compliance trend across runs becomes legible. See docs/agdr/AgDR-0019-audit-artefact-persistence.md for the schema rationale.

Resolve project name + score + verdict

<project-name> is the project's registered name in apexyard.projects.yaml. If unregistered, use the basename of the project path and tell the operator to /handover it.

Compute the headline score: score = max(0, 100 - 25*critical - 10*high - 3*medium - 1*low). Compute the verdict by worst-severity: critical/high → fail, medium → conditional, low/none → pass. The legacy "GDPR readiness" three-state vocabulary maps cleanly: NOT READY → fail, PARTIAL → conditional, READY → pass.

Persist + render

source "$(git rev-parse --show-toplevel)/.claude/hooks/_lib-audit-history.sh"

# Lowercase severity in the payload — the lib expects critical/high/medium/low/info.
payload=$(mktemp); cat > "$payload" <<'EOF'
{
  "schema_version": 1,
  "findings": [
    {"id": "C1", "severity": "high",   "status": "open", "summary": "No cookie consent UI; analytics fires before opt-in"},
    {"id": "C3", "severity": "medium", "status": "open", "summary": "Missing DPA with analytics vendor"},
    {"id": "C4", "severity": "high",   "status": "open", "summary": "No self-serve right-to-deletion endpoint"}
  ]
}
EOF

# Body: per templates/audits/compliance-check.md
body=$(mktemp); cat > "$body" <<'EOF'
... (filled-in body — findings table + Regulatory exposure + Recommended priority) ...
EOF

ts=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
audit_run_persist "<project-name>" "compliance-check" "$ts" "fail" 55  < 
 -f  

audit_render_trend   5

Opt-in commit (history-tracked marker)

touch projects/<name>/audits/compliance-check/.audit-history-tracked

The MD artefacts are committed regardless; the marker controls whether the per-run JSON files are.

Rules

1. Not legal advice. This audit identifies technical gaps. The user should consult a lawyer for legal compliance.
2. Region-aware. If the project doesn't target EU/UK users, some GDPR requirements may not apply. Ask the user about target markets if unclear.
3. Auto-PASS for non-user-facing projects. Internal tools, CLIs, and libraries without user data collection don't need this.
4. Check third-party SDKs. Analytics, error tracking, and ad SDKs often set cookies and process data — they need consent too.
5. Always persist via the lib. The persist step runs regardless of opt-in commit state. The marker only controls whether the JSON is committed.
6. Severity vocabulary in the JSON is lowercase. The lib's stats.by_severity derivation expects critical/high/medium/low/info. The human-readable findings table can use whatever capitalisation reads best.

---

Part of ApexYard — multi-project SDLC framework for Claude Code · MIT.