Skip to content

Latest commit

 

History

History
11 lines (8 loc) · 617 Bytes

SECURITY.md

File metadata and controls

11 lines (8 loc) · 617 Bytes

Working safely with skill packages

Benchmark packages are untrusted inputs, including their Markdown instructions.

  • Execute candidates and tests in isolated, disposable environments.
  • Restrict network access and expose only required task workspaces and fixtures.
  • Keep model-provider credentials outside task workspaces and logs.
  • Do not supply state labels, archived results, or private evaluators to the model.
  • Loading, hashing, and materializing inputs do not execute package code.

Report security concerns privately to the maintainers. Do not put credentials or private data in an issue or reproducer.