The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill methodology involves processing untrusted data from remote targets, creating a vulnerability surface.
Ingestion points: Data is ingested from network daemons, thick clients, and remote services via remote() and recvuntil functions in pwntools scripts as described in SKILL.md.
Boundary markers: The instructions do not include boundary markers or delimiters to separate untrusted external data from the agent's internal instruction context.
Capability inventory: The skill facilitates powerful capabilities such as network communication, binary exploitation, and shell access through the use of pwntools and referenced reverse shell payloads.
Sanitization: There are no instructions for sanitizing or validating service responses before they are processed by the agent's logic.
[DYNAMIC_EXECUTION]: The skill methodology involves generating and executing Python scripts at runtime.
Mechanism: The agent is instructed to build repeatable exploitation scripts using the pwntools library to interact with processes and remote targets.
Risk: While standard for exploit development, dynamic generation of code allows for the potential execution of logic derived from the analysis process.

exploit-memory-corruption