Skip to content

Security: hinanohart/polyphonic-eval

SECURITY.md

Security Policy

Supported versions

Version Supported
0.1.x
< 0.1

Reporting a vulnerability

If you discover a security vulnerability:

  1. Do not open a public GitHub issue.
  2. Open a GitHub Security Advisory: https://github.com/hinanohart/polyphonic-eval/security/advisories/new.

You should receive an initial response within 7 days. We will work with you to assess and disclose responsibly.

Threat model (initial)

This library:

  • does not make outbound network calls at import-time or in aggregate(). The default embedder (sentence-transformers) downloads model weights on first use; pin or pre-cache if your environment forbids that.
  • accepts arbitrary rationale strings from caller-supplied JudgeVerdict values. These are passed to embedders (numpy array path only). They are not eval'd, deserialized into Python objects, or used as format strings.
  • writes nothing to disk except via polyphonic_eval.io.jsonl when explicitly invoked by the caller.

Dependency security is monitored via Dependabot (configured in .github/dependabot.yml) and pip-audit recommended for downstream users.

There aren't any published security advisories