nemoclaw-deploy-remote

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). The SKILL.md explicitly wires the sandboxed OpenClaw agent to Telegram (steps on creating a bot, providing TELEGRAM_BOT_TOKEN, onboarding, and the instruction “After the sandbox is running, send a message to your bot in Telegram”), meaning the agent will ingest untrusted, user-generated Telegram messages that could influence its actions.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill instructs installing system software (Docker, NVIDIA Container Toolkit), starting host services, and baking/running sandbox images on the host—actions that modify system files and require elevated (sudo/system) privileges, so it pushes the agent toward changing machine state.