The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The SKILL.md file contains instructions for installing the 'GEF' (GDB Enhanced Features) plugin by downloading a remote script from gef.blah.cat and piping it directly into bash. This pattern involves executing unverified code from an external domain without integrity checks.
[DYNAMIC_EXECUTION]: The skill demonstrates using the pwntools library to dynamically generate machine-executable shellcode at runtime (e.g., shellcraft.sh()) and execute it using the process() function. While this is the intended purpose of the skill, it involves runtime code generation and execution.
[COMMAND_EXECUTION]: The scripts/agent.py script executes system utilities such as checksec and ROPgadget using the subprocess module to analyze binary files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted ELF binaries, which could be crafted to influence agent behavior or exploit the analysis tools.
Ingestion points: scripts/agent.py accepts binary paths via command-line arguments and loads them using the pwntools.ELF class.
Boundary markers: None identified in the provided instructions or scripts.
Capability inventory: scripts/agent.py executes system commands via subprocess.run; SKILL.md examples use pwntools to spawn and interact with processes.
Sanitization: No input validation or metadata sanitization is performed on the analyzed binaries before they are processed.

performing-binary-exploitation-analysis