Skip to content

v1.5.0

Latest

Choose a tag to compare

@x1xhlol x1xhlol released this 26 Sep 17:53

Changes since 1.4.0

  • Verdicts: a scan reports secure only when every check was graded. Otherwise the verdict is inconclusive and the score is 0.
  • Exit codes (breaking for scripts that match on 1): the CLI exits 1 on findings and 2 on errors or inconclusive scans. Errors used to exit 1.
  • Result fields: ScanResult has new coverage and models fields.
  • Endpoints: --base-url (or OPENAI_BASE_URL) points the scanner at any OpenAI-compatible endpoint.
  • Defaults: options you leave out, including empty model flags such as an unset "$VAR", take the engine defaults. The banner prints the models that actually run.
  • Library defaults: runSecurityScan turns on the inspector (with per-turn fingerprinting) and the orchestrator (with adaptive temperature), matching the CLI. This adds one inspector call per extraction turn.
  • README: this package is a source-available (FSL-1.1-Apache-2.0) prompt scanner that runs locally with your own keys. Hosted ZeroLeaks red-teams agents and has no free plan.