The Agent Skills Directory

[SAFE]: The skill provides documentation and routing prompts for NemoClaw, an NVIDIA-related agentic AI assistant framework. A thorough analysis of all files across 10 threat categories indicates that the content is benign and designed with a strong emphasis on security best practices, including credential isolation and sandbox hardening.
[PROMPT_INJECTION]: A static analysis flag regarding instruction overrides in references/nemoclaw-user-configure-security/references/openclaw-controls.md was identified as a false positive. The content is descriptive text explaining how the framework's security features detect prompt injection (e.g., 'ignore all previous instructions' patterns) rather than an attempt to execute such an attack.
[EXTERNAL_DOWNLOADS]: The documentation includes installation and maintenance commands that fetch scripts from official and well-known domains, such as nvidia.com and ollama.com. These downloads are recognized as safe vendor-provided resources and align with the primary purpose of the skill.
[COMMAND_EXECUTION]: The troubleshooting and setup guides include commands utilizing sudo, kill, and rm. These are standard administrative actions intended for host-side configuration (e.g., managing Docker groups or resolving port conflicts) and are appropriate within the context of managing an AI agent sandbox environment.

nemoclaw-user