The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes unvalidated data from network appliance responses, such as service banners, NTLM Type-2 challenges, and IKE handshake packets. This data could be manipulated by a target system to influence agent behavior.
Ingestion points: External data enters the context via tools like ike_enum.py, appliance_version_infer.py, and ntlm_decode.py.
Boundary markers: No explicit delimiters are specified to isolate untrusted network data from the instruction set.
Capability inventory: The skill directs the agent to execute various Python scripts and network probing tools.
Sanitization: The methodology does not describe steps to sanitize or filter network-derived strings before they are processed.
[EXTERNAL_DOWNLOADS]: The skill includes tools/nvd-lookup.py to query vulnerability data from the National Vulnerability Database (NIST), which is a well-known service.
[COMMAND_EXECUTION]: The skill workflow involves executing several local Python scripts, such as ike_enum.py, checkpoint_sic_opsec.py, and perimeter_forensics.py, to perform its intended security analysis functions.

network-appliance-offensive