The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides automated shell commands and Bash loops designed to perform active exploitation. For example, it includes a script that launches concurrent background curl processes to exploit race conditions in coupon redemption and payment systems.
[COMMAND_EXECUTION]: It provides detailed instructions for bypassing security controls and financial logic, such as manipulating price parameters, bypassing email verification flows, and accessing premium features without authorization.
[INDIRECT_PROMPT_INJECTION]: The skill reads and processes untrusted data from log.txt (captured network traffic) without implementing boundary markers or sanitization, creating a surface where malicious payloads in the traffic could influence the agent's behavior.
Ingestion points: Reads log.txt via grep commands in SKILL.md.
Boundary markers: Absent; traffic data is processed directly.
Capability inventory: Shell execution capabilities using grep and curl are present across the methodology.
Sanitization: No sanitization or escaping is performed on the data captured from the network logs before it is used in command templates or displayed.

mitm-find-bizlogic