The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The code examples in SKILL.md and references/examples.md demonstrate the use of pickle.load() for caching molecular embeddings. The pickle module is known for unsafe deserialization, where loading a maliciously crafted file can result in the execution of arbitrary code on the host system.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection (Category 8) due to its core function of processing external data.
Ingestion points: Untrusted molecular data in the form of SMILES strings are ingested into MoleculeTransformer and PretrainedMolTransformer instances across all referenced code examples.
Boundary markers: There are no instructions or delimiters provided to the agent to treat the ingested SMILES data as untrusted or to ignore any potential embedded instructions.
Capability inventory: The skill utilizes parallel processing via n_jobs (invoking subprocesses/multiprocessing), performs file system writes via to_state_yaml_file, and performs network-based model loading through its integration with HuggingFace and other repositories.
Sanitization: While the skill describes chemical standardization and salt removal using datamol, these processes are intended for data quality and do not sanitize against adversarial prompt injection techniques.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to install the molfeat library and various optional extras (such as dgl, graphormer, and transformer) from the PyPI package registry. It also references official project documentation and repositories hosted on GitHub and datamol.io.

molfeat