dsr-automation
Use when automating data subject requests: access, erasure, portability, and rectification at scale. Covers request intake and identity verification, orchestration across datastores, processor propaga
SKILL.mdskills/privacy/dsr-automation/SKILL.md
GitHubData subject requests arrive by the hundred: "send me my data", "delete everything". Manual handling does not scale and misses stores. This skill covers building the automation: verified intake, a fan out orchestrator that hits every datastore, propagation to processors, and SLA tracking that proves compliance.
Quick Workflow
- Build verified intake: request types, identity proofing proportionate to the data's sensitivity, and fraud resistance.
- Maintain the datastore inventory from the privacy registry; the orchestrator fans out to every entry.
- Implement per store handlers: export, delete, rectify, and restrict, each idempotent.
- Propagate to processors with receipts; retry failures visibly.
- Handle conflicts: legal holds, other people's data in the export, and backup windows.
- Track SLAs: deadlines per regulation, escalation before breach.
- Log everything for the regulator: what was done, when, where.
- Test with synthetic subjects quarterly, including the nasty cases.
1. Orchestrator sketch
#!/usr/bin/env python3
"""DSR orchestrator: fans a request out to registered store handlers."""
import datetime
HANDLERS = {} # store name -> handler object with export/delete methods
def register(name):
def deco(fn):
HANDLERS[name] = fn
return fn
return deco
@register("postgres_users")
def pg_users(action, subject_id):
return {"store": "postgres_users", "action": action,
"subject": subject_id, "status": "done",
"at": datetime.datetime.now(datetime.timezone.utc).isoformat()}
def run_dsr(kind, subject_id):
log = []
for name, fn in HANDLERS.items():
try:
log.append(fn(kind, subject_id))
except Exception as e: # noqa: BLE001 - one store failing must not stop others
log.append({"store": name, "status": "FAILED", "error": str(e)})
return log
print(run_dsr("export", "user_123"))
# Production: persist the log, retry FAILED, alert on repeated failure
2. Identity verification and conflicts
# Verification proportional to risk:
# - export of basic profile: logged in session is enough
# - export of sensitive data: step up auth
# - erasure: confirm explicitly, warn about irreversibility
# Fraud watch: bursts of requests from one source, requests for others
echo "log verification method per request; regulators ask"
# Conflict handling:
# - legal hold: pause erasure for held data, document the hold
# - mixed data (photos with other people): redact third parties in exports
# - backups: erase from live now, purge backups on their cycle, document it
psql "$DB_URL" -c "SELECT * FROM legal_holds WHERE subject_id='user_123';"
Edge cases
- The subject cannot be verified. You may refuse, but you must tell them why and what would suffice.
- Processors do not confirm. Escalate per contract; repeated failure is a vendor management issue.
- Request volume spikes (campaign driven). Queue with priority by deadline, never silently drop.
- The export format must be portable (JSON/CSV), not a PDF screenshot.
Escalation
- SLA breach imminent: manual triage of the oldest requests while automation catches up.
- Systemic handler failure: fix the handler, then re run affected requests; notify subjects if the delay breaches the deadline.
- Regulator inquiry: the per request log is your evidence; keep it complete and tamper evident.
Detection and defender view
Detect DSR automation gaps with synthetic subjects: submit export, rectify, and erase requests and verify every store. Monitor handler failure rates, SLA burn down, and verification fraud signals. The durable design treats DSR as a product feature with owners and tests, not a legal chore done in spreadsheets.
Key references
- GDPR Articles 12-22: rights and modalities.
- EDPB guidance on data subject rights.
- Related skills: gdpr-engineering, consent-management.