The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted network traffic and packet capture files, creating a potential surface for indirect prompt injection if the captured data contains malicious instructions targeting the agent.
Ingestion points: Reading network traffic or capture files via wireshark, tcpdump, tshark, and scapy.rdpcap as described in SKILL.md.
Boundary markers: No explicit instructions are provided to the agent to distinguish between the data being analyzed and its own control logic.
Capability inventory: The skill involves active packet sending (scapy.send), system-level traffic capture (tcpdump), and man-in-the-middle interception (mitmproxy).
Sanitization: There is no evidence of sanitization or filtering of captured packet content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill includes numerous examples of shell commands for network capture and interception that interact with system interfaces.
Evidence: Bash snippets in SKILL.md provide commands for wireshark, tcpdump, and tshark on interface eth0.
Evidence: Python code blocks demonstrate active network interaction, such as replaying and modifying packets with scapy and fuzzing protocols with boofuzz.
[EXTERNAL_DOWNLOADS]: The skill documentation references external libraries commonly used for security research.
Evidence: Python examples in SKILL.md rely on the scapy and boofuzz packages for advanced protocol analysis and testing.

protocol-reverse-engineering