Skip to content

Security

By the end of this page you know what redfish protects, the one thing it never does, and how to report a vulnerability.

This is the core of the platform. The simulated login page records no input: redfish measures that a target clicked, never what they typed. A change that stores what a target types will not be merged. Anything that would let the platform capture credentials is treated as a critical vulnerability.

In production the operator dashboard and the public tracking links run on two different host names, set through DASHBOARD_HOST and TRACKING_HOST. The tracking path /t/* is served only on the tracking host, the operator application only on the dashboard host, and any cross-host or unknown-host request gets a 404. This keeps the operator interface off the host that targets interact with. See Configuration.

redfish sends a strict policy: default-src 'self', no external origin, and script-src 'self' with a per-response nonce on the inline scripts. It never uses 'unsafe-inline' for scripts. Your reverse proxy should not log request bodies, so that nothing a target might type is written down, and should not log the Cookie or Authorization headers. See Configuration.

A target who clicks lands on an awareness page that explains the warning signs. The goal is to teach, not to trick people into giving up information. See Scenarios.

Sensitive actions are rate limited. Personal data is kept only as long as needed: expired targets are anonymized daily. See Operations.

Use GitHub private vulnerability reporting: open an advisory at https://github.com/selfhostr/redfish/security/advisories/new, as described in SECURITY.md. Never open a public issue for a vulnerability. Only main and the latest tag are supported.

Of special interest are findings that would let the platform capture credentials, bypass the host isolation between dashboard and tracking, leak secrets stored in the database, or read data across organizations.