The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system utilities including tc (traffic control), iperf3, ping, and network attack tools like arpspoof and bettercap to simulate network degradation and traffic interception. These commands are executed via shell scripts in SKILL.md and through the subprocess module in scripts/agent.py.
[PRIVILEGE_ESCALATION]: Root access via sudo is required and explicitly used throughout the instructions and scripts to modify kernel-level network queuing disciplines (qdiscs), modify iptables rules, and perform packet crafting with scapy.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by consuming and parsing output from external network tools.
Ingestion points: The measure_baseline function in scripts/agent.py parses JSON output from the iperf3 command. Additionally, SKILL.md suggests analyzing Zeek connection logs and NetFlow data.
Boundary markers: No explicit boundary markers or instructions to ignore embedded content are used when processing the external tool outputs.
Capability inventory: The agent has the ability to execute arbitrary system commands via subprocess.run (found in scripts/agent.py) and perform network operations using the scapy library.
Sanitization: Basic JSON parsing is performed on iperf3 output, but no further validation or sanitization of the values is evident before the data is used in report generation.

performing-bandwidth-throttling-attack-simulation