Open a GitHub issue for anything that isn't sensitive. For anything you'd rather not post publicly (a real exploit against the live site, a credential leak, an auth bypass), email usher@rndsoftwaregroup.com instead of filing an issue.

• No secret is ever committed. .env is gitignored; .env.example holds no real values. If you find one in the history anyway, treat it as compromised and say so in the report — it doesn't matter whether it's still "valid," it needs rotating.
• No secret is baked into a Docker image layer. Every credential the app needs (D1, R2, the tunnel token) arrives as a runtime environment variable injected by the CI deploy step, never an ENV/ARG in the Dockerfile and never a file written to bulky-server's disk. docker history on either published image should show no secret material — if it ever does, that's a real finding, not a style nit.
• The server holds no application data. bulky-server has no volumes and runs the app container read_only. Sons live in D1, images live in R2. A compromised container has nothing local worth taking beyond whatever environment variables it was handed for that run.
• Base images are pinned by digest, not a movable tag — see the README's Deploy section for why and how to re-pin.
• Moderation is unproven on real explicit content. The CLIP-based NSFW gate has been checked for false positives (safe images wrongly rejected) but not against a real abuse benchmark — see the README's Moderation section. Don't treat nsfw_score < 0.5 as a guarantee.

This repository's Cloudflare, Tailscale, and R2 credentials were rotated multiple times during initial setup after being pasted into a chat transcript rather than set directly. If you're the one operating this going forward: prefer writing credentials straight into .env or a secret manager over typing them into any chat interface, including this one — a transcript is a durable record whether or not the conversation feels ephemeral.