The AI Catalog is a public, read‑only catalog with a small admin‑only write surface. Security is enforced in two layers.

• Rate limiting — 240 requests/min per IP on the API.
• Admin‑token guard — every write endpoint (create/update/delete) requires an admin token. Public reads are unauthenticated and unaffected.
• Input validation — all inputs are validated with Zod.
• Parameterized queries — database access uses Drizzle ORM with parameterized queries, so SQL injection, XSS and path traversal are not exploitable.

The site is served behind a reverse proxy that sets hardened response headers:

• Strict-Transport-Security — HSTS, 1 year, includeSubDomains; preload
• X-Frame-Options: DENY — anti‑clickjacking
• X-Content-Type-Options: nosniff
• Referrer-Policy: same-origin
• Content-Security-Policy — own‑origin only (fonts allow‑listed)
• Server / version banners removed (Server, X-Powered-By)

A reference header block is in deploy/security-headers.caddy. The full reverse‑proxy configuration (host and upstream details) is intentionally not committed.

An independent penetration test — SQL injection, XSS, command injection, path traversal, headers and data exposure — scored 10/10, zero critical vulnerabilities.

Please do not open a public issue for an undisclosed vulnerability. Open a private security advisory on this repository, or contact the maintainers. We appreciate responsible disclosure.