The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/agent.py script invokes the nmap utility using subprocess.check_output to perform vulnerability checks. While implemented securely using a command list, this allows the agent to interact with the local networking environment.
Evidence: subprocess.check_output(["nmap", "-p", "445", "--script", "smb-vuln-ms17-010", target_ip], ...) in scripts/agent.py.
[INDIRECT_PROMPT_INJECTION]: The vulnerability scanning logic in scripts/agent.py and scripts/process.py reads raw responses from remote SMB services. This data is incorporated into the agent's output and reports, potentially allowing a malicious host to provide crafted responses.
Ingestion points: Raw socket reads in scripts/agent.py and scripts/process.py using sock.recv(4096).
Boundary markers: The skill does not use explicit boundary markers or "ignore instructions" tags when processing or reporting the results of these network reads.
Capability inventory: The skill includes file system write capabilities (args.output) and external command execution (nmap).
Sanitization: There is no sanitization or escaping of the network-derived data beyond checking for specific binary SMB header patterns.

exploiting-ms17-010-eternalblue-vulnerability