The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted ciphertext strings and files, which serves as a potential vector for indirect prompt injection if the resulting plaintext contains instructions that influence the agent's future actions.
Ingestion points: CLI arguments and file read operations in scripts like classic/vigenere.py, classic/xor_single_byte.py, and rsa_tool/small_e.py.
Boundary markers: The scripts lack specific delimiters or instructions to ignore potential commands embedded within the data being analyzed.
Capability inventory: Perform modular arithmetic, decrypt RSA ciphertexts, conduct frequency analysis, and perform local file operations (read/write).
Sanitization: Standard error handling for data conversion and command-line argument parsing is present, but no sanitization for instructional content within ciphertexts is implemented.
[EXTERNAL_DOWNLOADS]: The skill's documentation points to external code repositories for additional tools.
Evidence: README.md and SKILL.md reference cloning the RsaCtfTool project from non-whitelisted GitHub repositories (RanSommer and Ganapati).
[REMOTE_CODE_EXECUTION]: The skill provides instructions to execute code downloaded from external sources.
Evidence: Instructions in README.md guide the user/agent to clone a repository and execute the RsaCtfTool.py script locally.

crypto-tools