This tool is currently in early-stage development. Security fixes are best-effort.

If you discover a security issue, do not open a public issue. Contact the maintainers privately.

• Treat suites, predictions, and model outputs as untrusted input.
• Run in a sandboxed CI environment when evaluating untrusted content.