mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-18 15:29:36 +00:00
1.6 KiB
1.6 KiB
Security & PDPL Checklist — Dealix
Data minimization
- Collect only fields required for the revenue workflow.
- Separate PII from embedding payloads unless explicitly allowed and documented.
Consent & suppression
- Consent ledger for marketing / WhatsApp outreach.
- Opt-out / suppression list enforced before any campaign or message draft batch.
Contactability
- Channel-specific rules (email vs WhatsApp vs LinkedIn).
- No cold WhatsApp; LinkedIn DMs not automated.
Audit
- Audit logs for outbound drafts, approvals, and agent tool calls.
- Correlation IDs on API requests (
RequestIDMiddleware).
Secrets & embeddings
- No secret indexing — block
sk-, private keys, bearer tokens (seelooks_like_secret,should_block_embedding). - No tokens in embeddings — redact before chunk upsert.
Supabase
- Service role key only on server-side runtimes (Railway/Render/Fly/VPS).
- RLS enabled; policies reviewed before any client exposure.
- Retention policy for logs, embeddings refresh, and deletion requests.
Data subject rights
- Export process documented.
- Delete process documented (including vectors and CRM mirrors).
Outbound messaging
- Approval-required for Gmail send, WhatsApp send, calendar create.
- Admin-only tools for project memory bulk reindex in production.
Approval-required messaging (product)
- Personal Operator endpoints return
approval_required: trueon drafts by design. - Calendar route documents Arabic + English notes that real creation needs explicit approval.