deepstream-setup-on-wsl

Pass

Audited by Gen Agent Trust Hub on Oct 31, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes shell and PowerShell commands on both the local controller and remote Windows targets. Remote execution is managed via secure WinRM (HTTPS) and SSH with strict host-key verification. A user-supplied command feature implements a restrictive whitelist of allowed executables (e.g., gst-launch-1.0, nvidia-smi) and explicitly rejects shell operators to prevent command injection.
  • [EXTERNAL_DOWNLOADS]: Fetches required software components from well-known sources, including Microsoft's GitHub repository, Docker's official repository, and NVIDIA's developer domains. High-risk downloads, such as the WSL MSI package, are verified against hardcoded SHA-256 hashes and Authenticode signatures before installation.
  • [CREDENTIALS_SAFE]: Implements a private credential management system using configure_target.py. It explicitly forbids passing passwords through the agent chat, instead prompting the user in a local interactive terminal and storing secrets in 0600-mode files within a private, user-owned directory (~/.deepstream-wsl/).
  • [DATA_EXFILTRATION]: While the skill transfers log files from the target machine back to the controller, it uses a hardened tar extraction method to prevent path traversal attacks. Data transfer occurs over the established secure WinRM or SSH channels and requires explicit user confirmation.
  • [SAFE]: The skill demonstrates extensive defensive programming, including strict input validation (regex checks on hostnames and image tags), ownership and permission checks on private directories, and explicit user confirmation gates before any state-changing operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 31, 2026, 09:41 PM