xiao-esp32s3-mqtt-dashboard

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill instructs the agent to ask the user for their WiFi SSID/password and to literally replace placeholders in the firmware template (e.g., WIFI_PASS) with those values, forcing the LLM to output secrets verbatim into generated code.

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

• Third-party content exposure detected (high risk: 0.85). Yes—runtime flow includes browser dashboard JS (assets/dashboard_web/app.js) connecting to an MQTT broker over WebSockets and rendering MQTT message payloads (outsider-authored topic/payload text from other devices/users) into the page via devices.set(...) and render(); while the code does not directly send this text to an LLM, it is still outsider-authored free text ingested into the client context.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The dashboard's index.html loads and executes the remote script https://unpkg.com/mqtt/dist/mqtt.min.js at runtime (app.js depends on the global mqtt object), so the skill fetches and runs external code the dashboard requires.

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 explicitly instructs the user/agent to copy and run Windows setup scripts with elevation (Start-Process ... -Verb RunAs), modify Mosquitto service/config under C:\Program Files and add firewall rules—actions that change system state and require admin privileges.