compliance-check

Prevent the common mistake of shipping a SaaS with real user data before the minimum legal and trust foundations are in place. By reading the PRD, architecture, and data-model docs, the skill builds an inventory of every data item collected or stored, flags which items are PII, traces where each item lives (Supabase, Stripe, Vercel, or a third party), and checks whether the app has the policy documents and code paths needed to honor GDPR data-subject rights (export and delete). It then scores each compliance item as PASS or GAP and produces a prioritized list of the open gaps to close before onboarding real users. Non-autonomous: drafts are presented and require explicit approval before any file is written.

Phases

1. Load context — read docs/specs/prd.md and docs/specs/architecture.md; also read docs/specs/data-model.md if it exists. If the PRD is missing, stop and direct the user to /write-prd. Extract what data the app collects, from whom, and for what purpose.
2. Data inventory — build a table of every data item collected or stored, with columns: Item, Is PII (yes/no), Storage location (Supabase / Stripe / Vercel / other), Retention period (known or TBD). Present the table to the user for corrections and additions before proceeding.
3. Policies & rights — check for the presence of: (a) a privacy policy stub and Terms of Service; (b) a DPA / subprocessors list covering at minimum Supabase, Stripe, and Vercel; (c) a data-export path that lets a user or tenant download their data; (d) a data-deletion path that removes or anonymizes PII on request; (e) cookie consent if any tracking cookies are used; (f) (SOC 2-lite) an audit trail — key events (authentication, billing changes, and destructive admin actions) are logged to a tamper-evident log or append-only audit_events table. For each item, mark PASS (exists and adequate), GAP (missing or incomplete), or N/A (not applicable). Under full review, confirm each item with the user; under lean, present the full checklist and confirm once; under solo, write then summarize.
4. Draft — fill .claude/templates/compliance.md into docs/specs/compliance.md with the data inventory table and the PASS/GAP checklist. Present the draft and await explicit approval before writing the file.
5. Verdict — list all open GAPs in priority order (blocking: items that affect real-user data handling; important: items needed before public launch; advisory: best-practice items). Make clear which GAPs must be closed before the app can handle real user data.