The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted cryptographic parameters and data (e.g., ciphertexts, public keys, and modulus values) provided by users or fetched from external CTF challenges across nearly all markdown files in the skill.\n
Ingestion points: Variable placeholders like '', '', and '' in 'SKILL.md', and input arguments to various attack functions in files like 'classic-ciphers.md', 'modern-ciphers.md', and 'rsa-attacks.md'.\n
Boundary markers: Absent; the skill does not use specific delimiters to encapsulate challenge-provided data within the agent's prompt context.\n
Capability inventory: The skill can execute complex Python scripts using the 'Bash' tool and interact with remote servers via 'WebFetch'.\n
Sanitization: Absent; external data is processed directly as mathematical integers or byte sequences without specific filtering for prompt injection patterns.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions to install several specialized libraries and tools from well-known public repositories.\n
Fetches Python packages including 'pycryptodome', 'z3-solver', 'sympy', 'gmpy2', 'hashpumpy', 'fpylll', and 'py_ecc' from standard registries.\n
Downloads well-known cryptanalysis tools like 'RsaCtfTool' and the 'crypto-attacks' library from their official GitHub repositories.\n- [DYNAMIC_EXECUTION]: In 'prng.md', the skill demonstrates the use of 'ctypes.CDLL("./libc.so.6")' to load a local shared library and 'eval()' on a local data file ('magic_data'). These techniques are used to synchronize state with challenge binaries and load precomputed mathematical matrices.\n- [COMMAND_EXECUTION]: The skill frequently uses the 'Bash' tool to run Python scripts and interact with system tools such as 'file', 'hashcat', 'tshark', and 'fastcol' to perform forensic and cryptanalytic tasks.

ctf-crypto