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=...)andSTRANDS_ISAAC_NUCLEUS_URLpoint 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).