Skip to content

Security

Found a vulnerability? Do not open a public GitHub issue.

Amazon Web Services (AWS) coordinates responsible disclosure of security vulnerabilities for this project.

Submit the issue privately to the AWS Vulnerability Disclosure Program:

More details: AWS Vulnerability Reporting.

The repo's SECURITY.md file is the canonical source.

Threat model notes

strands-robots-sim is a thin plugin layer over Isaac Sim. The bulk of the security surface — natural-language tool dispatch, mesh networking, HuggingFace trust_remote_code gating, IoT bridging — lives upstream in strands-labs/robots. File issues there for the agent / mesh / hardware paths.

What this repo adds to the threat model:

  • Isaac Sim Kit extensions. Loading USD / URDF assets from untrusted sources runs through Omniverse Kit's importer pipeline. Treat untrusted USD files like untrusted code.
  • Replicator output paths. replicate(output_dir=...) writes to arbitrary host paths. Validate the path against your dataset root in caller code; the engine does not sandbox it.
  • Nucleus URLs. IsaacConfig(nucleus_url=...) and STRANDS_ISAAC_NUCLEUS_URL point Kit at a Nucleus server for asset resolution. A malicious Nucleus can serve crafted USDs that abuse the importer. Pin to trusted servers.

Reporting checklist

When filing a private report, please include:

  • Repo + tag / commit SHA you tested against.
  • Repro snippet (Python script + which Isaac Sim version).
  • Impact statement: what does the bug let an attacker do?
  • Suggested fix (if any).