KimCad is a beta-stage, local-first tool. Please report security issues privately — use GitHub's private vulnerability reporting (Security → Report a vulnerability) rather than opening a public issue.

• The affected command, route, or workflow.
• Exact reproduction steps.
• Any generated files needed to reproduce the issue.
• Whether the issue involves generated code, printer credentials, local files, cloud API keys, or network exposure.

KimCad is designed to run on a trusted user's own Windows machine. The web server binds to loopback by default, generated CAD code is sandboxed and validated before slicing, and printer sends require explicit confirmation.

Session token (KC-26). State-changing requests carry a per-boot random token (issued by the server, injected into the page shell, returned as the X-KimCad-Session header; constant-time compared, 403 on mismatch). This is defense-in-depth against a drive-by cross-origin POST from a malicious web page — which can reach loopback but cannot read the same-origin token — and is deliberately not full CSRF protection (a single-user loopback app has no cookie session to forge) and not authentication. In particular it does not secure --allow-remote: any client that can load the page over HTTP reads the token, so a remotely-exposed instance is still unauthenticated and needs a separate auth layer (see below). Side-effecting GETs that can't carry the token (the lazy STEP build, the health re-probe) refuse a cross-origin request via Sec-Fetch-Site. Full detail: docs/api.md (Security model).

Do not expose kimcad web on an untrusted network without a separate authentication layer. Real printer validation is intentionally deferred until the post-Stage-11 hardware phase.