Scientific type safety validates object flow within the registered workflow. It is distinct from programming-language type soundness or an operating-system isolation boundary.

The optional code-acting baselines execute model-proposed Python through their documented resource controls. Run model-backed services in a trusted local environment and bind them to localhost unless you provide appropriate deployment controls.

Keep model-provider credentials in local environment configuration. Do not include keys in issue reports. Report security concerns privately to the repository maintainer through GitHub's private reporting facility when available; include the affected revision and a minimal reproducer without real credentials.