serial-shell

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

• Third-party content exposure detected (high risk: 0.85). Outsider text can enter the LLM context via the device’s serial output: shell_proxy.py reads arbitrary bytes from the connected device (ser.read(...)) and prints/optionally logs them, and the skill’s runtime then returns those printed/logged contents to the agent; this is not authored by the operating user.