The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill uses eval() in references/prng.md to load data from a local file, which can execute arbitrary code if the file content is untrusted.
[DYNAMIC_EXECUTION]: Use of ctypes.CDLL to dynamically load local shared libraries, such as libc.so.6, into the agent's process.
[COMMAND_EXECUTION]: Several reference files contain code that uses subprocess.run, subprocess.check_output, and os.system to execute external security tools like fastcoll, crchack, and roca-detect.
[DATA_EXFILTRATION]: The skill performs network operations using requests and the pwn library's remote function to interact with external challenge servers and oracles.
[EXTERNAL_DOWNLOADS]: Instructions are provided to download and install various third-party security tools and libraries from repositories on GitHub and GitLab.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from cryptography challenges, including ciphertexts and transcripts, which represents a potential injection surface. This is combined with capabilities such as command execution and network operations.

crypto-ctf