counterscreen is research software. It produces design candidates and statistical claims; it does not handle credentials, personal data, or patient data, and it exposes no network service.
Two areas are worth a security eye regardless:
- Third-party model execution. The
generate,sequenceandfoldpackages drive external binaries and model weights as subprocesses. A malicious weights file or a crafted structure input is an untrusted input to someone else's parser. - Provenance integrity. Results under
campaigns/*/results/are evidence and are append-only. A defect that lets a recorded measurement be silently rewritten is a correctness and integrity bug, and we treat it as the more serious of the two.
Report privately through GitHub's advisory form:
https://github.com/mbote-droid/counterscreen/security/advisories/new
Please do not open a public issue for anything exploitable. Include what you did, what happened, and the commit you were on. Expect an acknowledgement within a week; this is a small project and there is no on-call rota.
Pre-1.0. Only the tip of the default branch is supported. No backports.
- A model producing a bad design. That is a scientific result, and the control panel exists to catch it.
- Needing a GPU, large weights, or licensed third-party software to run the generative stages.
- Resource exhaustion from deliberately oversized structure inputs run locally by the operator.