The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install numerous third-party tools and libraries essential for cryptanalysis from public repositories and registries.
Installs Python packages via pip, including pycryptodome, z3-solver, sympy, gmpy2, hashpumpy, fpylll, py_ecc, and galois.
Guides the user to clone the RsaCtfTool repository from https://github.com/RsaCtfTool/RsaCtfTool.
Guides the user to clone the crypto-attacks repository from https://github.com/jvdsn/crypto-attacks.
[COMMAND_EXECUTION]: The provided code snippets and one-liners involve the execution of system commands and interfacing with system-level resources to perform analysis.
Uses subprocess.run to invoke tools like file and hashcat for format identification and password cracking.
Uses the ctypes library to load libc.so.6 to synchronize and predict random number generation outputs.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from CTF challenges, creating an inherent surface for indirect instruction processing.
Ingestion points: Cryptographic parameters (moduli, exponents, ciphertexts) and files are read from the local filesystem or fetched via network tools to be processed by analysis scripts.
Boundary markers: The analysis snippets do not consistently implement boundary markers or instructions to ignore potential payloads embedded within the cryptographic data being analyzed.
Capability inventory: The skill has access to extensive capabilities including shell command execution (Bash), filesystem access (Read, Write, Edit), and network access (WebFetch, WebSearch).
Sanitization: Cryptographic inputs are generally handled as raw integers or byte sequences without specific sanitization for potential prompt injection payloads.

ctf-crypto