Skip to content

Security: seismael/openharness

Security

SECURITY.md

Security policy

Supported versions

Version Supported
0.2.x Yes
0.1.x No

Reporting a vulnerability

Report vulnerabilities privately through GitHub Security Advisories. Do not open a public issue for a suspected vulnerability. Include the version, platform, a minimal reproduction, and the impact you observed. Expect an initial response within seven days.

Design boundaries

OpenHarness is a local, deterministic tool. It makes no network calls, sends no telemetry, and needs no credentials. That removes whole classes of risk, and it also means the project does not claim to be a runtime enforcement layer. Runtime policy enforcement belongs to the host; OpenHarness evaluates and emits.

What to treat as untrusted

  • Declarations. harness.json, policy.json, cases.json, and requirements.json are inputs. Review them like code, because they decide what gets measured and what a policy permits.
  • Regular expressions. Policy rules use command_regex. A pathological pattern can be slow. Keep patterns anchored and simple; the evaluator applies last-match-wins over a small rule set.
  • Commands. openharness cases and openharness report --execute run the commands declared in a case file, through the shell, in the selected working directory. Run them only from a case file you trust, exactly as you would run a CI script. openharness report without --execute never runs anything.
  • Paths. init and lock write only to explicit paths under the repository root. They do not follow symlinks out of the repository by design; the release checker rejects symlinks in published bundles.

Supply chain

Releases are built from tagged commits in GitHub Actions, published to PyPI with OIDC trusted publishing (no long-lived token), and attested with SLSA build provenance. Verify with gh attestation verify <wheel> --repo seismael/openharness.

There aren't any published security advisories