version: "1.0.1" name: security-hardening description: "Security architecture: authentication, authorization, RLS policies, CSP, input validation, API security. Use when implementing auth flows, writing RLS policies, configuring CSP/headers, validating inputs, or auditing security. Trigger terms: RLS, CSP, Server Actions, Zod, auth flow"

Security Hardening

Authentication

Resolve the auth library via the database capability slot in the skill matrix.

• Every sign in/up/out goes through a Server Action (POST-only → automatic CSRF protection).
• Session refresh in middleware updateSession() with HTTP-only cookies; protected routes checked in middleware.
• User roles live in profiles.roles TEXT[].

CSP

Least privilege; whitelist only required external domains per directive (project-specific, see deployment customization). 'unsafe-inline'/'unsafe-eval' may be needed in dev — use nonces/hashes in production. Validate shipped headers with curl -I against the preview URL.

RLS

SQL examples and role system: see the database skill (authoritative source for RLS).

• ALTER TABLE x ENABLE ROW LEVEL SECURITY; on every table — default-deny, explicit-allow.
• auth.uid() for auth checks, EXISTS subqueries for role checks.
• Never disable RLS in production; never rely on client-side authorization alone.
• CI gate: assert SELECT relrowsecurity FROM pg_class WHERE relname = 'your_table' is true, plus a positive/negative row-visibility test (other role must read 0 rows). Block merges on failure.

API Security

Cron routes: require authorization: Bearer ${process.env.CRON_SECRET}, else return 401. Generate with openssl rand -hex 32; rotate quarterly.

Zod schema validation on every Server Action and route handler before any DB operation; React Hook Form client-side.

Cross-reference: the api-patterns skill for Server Action patterns; the session-checkpoints skill for checkpointing security-sensitive work.