The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to process and analyze data from external, untrusted IoT protocol sources, which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: SKILL.md (processing network capture files via tshark, subscribing to MQTT topics via mosquitto_sub, and reading BLE/NFC data).
Boundary markers: Absent. The instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the processed data.
Capability inventory: The skill utilizes subprocess calls to CLI tools such as tshark, mosquitto_pub, coap-client, and ubertooth-btle across various operational steps.
Sanitization: Absent. No explicit sanitization or validation of the protocol data is described.
[EXTERNAL_DOWNLOADS]: The skill recommends installing several security-related packages from official registries and the Python Package Index (PyPI).
Evidence: Instructions include pip install scapy, pip install aiocoap, and pip3 install killerbee.
[COMMAND_EXECUTION]: The skill provides numerous command-line examples for system configuration and protocol interaction.
Evidence: Commands such as sudo dpkg-reconfigure wireshark-common, sudo ./kb_install_firmware, and active protocol interactions like mosquitto_pub and coap-client are documented.

re-iot-proto