The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill fetches and executes scripts from NVIDIA's official domain and GitHub repository by piping output from curl directly into bash and sh.
[COMMAND_EXECUTION]: Requires administrative privileges (sudo) to install system packages like docker.io and nodejs, and to modify system configuration files such as /etc/docker/daemon.json and /etc/systemd/system/.
[COMMAND_EXECUTION]: Establishes persistence by creating a custom systemd service (nemoclaw-forward.service) and a local keepalive script (nemoclaw-keepalive.sh) to maintain background port forwarding.
[EXTERNAL_DOWNLOADS]: Downloads software components from external sources including NodeSource and NVIDIA-affiliated repositories.
[PROMPT_INJECTION]: The skill configures an AI agent environment that processes untrusted workspace data while possessing broad execution capabilities.
Ingestion points: Workspace profile and behavior files (USER.md, SOUL.md, TOOLS.md) and custom skills defined in the sandbox environment.
Boundary markers: Absent; no explicit delimiters or instructions are provided to the agent to differentiate between static data and active instructions.
Capability inventory: The agent is configured with high-impact tools including exec, read, and write access within its environment.
Sanitization: Absent; content is written directly to workspace and skill files without validation or sanitization of external content.

nemoclaw-setup