nemoclaw-deploy-remote

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill documents the integration of Telegram bot messaging, which introduces an indirect prompt injection surface when the agent processes untrusted user input.
  • Ingestion points: User messages delivered via the Telegram bot token configuration in SKILL.md.
  • Boundary markers: The documentation suggests using an allowlist (TELEGRAM_ALLOWED_IDS) for access control, though it does not explicitly detail the use of delimiters for data isolation within the prompt.
  • Capability inventory: The agent can execute commands and manage network requests through OpenShell, with a terminal interface provided for approving requests.
  • Sanitization: Security hardening measures are detailed in references/sandbox-hardening.md, including the removal of build tools (gcc, g++, make) and network probes from the sandbox image to limit post-exploitation capabilities.- [EXTERNAL_DOWNLOADS]: The skill references the installation of the Brev CLI from brev.nvidia.com to manage remote GPU provisioning. This is an official vendor resource.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 04:30 AM