nvidia-nemoclaw

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches installation scripts and configuration from NVIDIA's official domain (nvidia.com) and GitHub repositories (github.com/NVIDIA).
  • [REMOTE_CODE_EXECUTION]: Executes a setup script directly from NVIDIA's official web server via a shell pipe.
  • [COMMAND_EXECUTION]: Utilizes system commands like git, npm, and docker to install dependencies, build the project, and manage containerized sandboxes.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external agent messages within a sandbox environment, creating a potential attack surface for malicious input to influence agent behavior. Security is managed through OpenShell policy enforcement including network and filesystem restrictions.
  • Ingestion points: User-provided messages processed by the openclaw agent or NemoClawClient (SKILL.md).
  • Boundary markers: Sandbox policies (Landlock, seccomp, netns) and OpenShell gateway monitoring (SKILL.md).
  • Capability inventory: File writes to /sandbox and /tmp, network egress to specified hosts, and shell command execution inside the sandbox (SKILL.md).
  • Sanitization: Policy-enforced egress blocking and interactive operator approvals via OpenShell TUI (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 08:37 PM
Security Audit — agent-trust-hub — nvidia-nemoclaw