The Agent Skills Directory

[COMMAND_EXECUTION]: The skill suggests using subprocess.run to execute external CLI tools such as mosquitto_pub. This pattern poses a risk of command injection if the command arguments or payloads are constructed from untrusted user input without adequate sanitization.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from external MQTT brokers.
Ingestion points: Data enters the agent context through MQTT message payloads in the on_message callback (e.g., message.payload.decode()).
Boundary markers: There are no delimiters or instructions provided to the agent to treat the incoming MQTT payload as untrusted data rather than instructions.
Capability inventory: The skill possesses capabilities for shell command execution (subprocess.run), file system interaction (JSON config reading), and network operations (client.publish).
Sanitization: The examples provided do not include any logic to sanitize, validate, or escape the content of the MQTT messages before they are processed or logged.

mqtt