The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's documentation and code examples promote the use of the Python pickle module for serializing and deserializing spectral data (e.g., load_from_pickle, save_as_pickle, scores.to_pickle). The pickle module is insecure because it can execute arbitrary Python code during the loading process. If an agent is directed to process a malicious .pkl file provided by an untrusted source, it could lead to remote code execution and system compromise.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting and processing metadata from various spectral data formats (mzML, MGF, MSP, JSON) and remote Universal Spectrum Identifier (USI) references. Maliciously crafted metadata, such as compound names or comments, could contain instructions intended to influence the agent's behavior. The instructions do not include specific boundary markers or sanitization steps to mitigate this risk.
[DATA_EXFILTRATION]: The skill performs network requests to external scientific databases to enrich spectral data. It uses derive_annotation_from_compound_name to query PubChem for chemical identifiers and load_from_usi to retrieve spectral data from metabolomics repositories such as GNPS. While these are legitimate features, they represent paths for external data communication.
[COMMAND_EXECUTION]: The skill includes instructions to install the required Python environment using the uv pip install matchms command.
[EXTERNAL_DOWNLOADS]: The skill requires downloading and installing several third-party libraries from the Python Package Index (PyPI), including matchms, numpy, pandas, and rdkit (via the [chemistry] optional dependency).

matchms