The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides shell commands for installing the molfeat library and its optional dependencies (e.g., molfeat[all], molfeat[transformer]) using uv pip install.- [EXTERNAL_DOWNLOADS]: The skill uses PretrainedMolTransformer and the ModelStore class to fetch and load pre-trained molecular models and embeddings from external remote repositories during runtime.- [REMOTE_CODE_EXECUTION]: The documentation includes a code example for caching embeddings that uses Python's pickle.load() and pickle.dump() methods. Deserializing data from untrusted files using pickle can lead to arbitrary code execution if a user or the agent loads a malicious cache file.- [PROMPT_INJECTION]: The skill processes untrusted molecular data (SMILES strings) as input for featurization. This creates an indirect prompt injection surface where malicious instructions could be embedded in molecular metadata or sequences, potentially influencing downstream agent decisions if the featurized output or errors are re-incorporated into the LLM context without sanitization.

molfeat