The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation in SKILL.md and references/importing_exporting.md demonstrates the use of load_from_pickle for handling preprocessed spectra. The Python pickle module is insecure against maliciously constructed data, and loading an untrusted pickle file can result in arbitrary code execution on the host machine.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes mass spectrometry data from external formats such as MGF, MSP, mzML, and JSON. These files contain metadata fields like compound names and comments that are ingested by the agent, creating a surface for indirect prompt injection where malicious instructions could be embedded in the scientific data.\n
Ingestion points: Spectrum loading functions in references/importing_exporting.md including load_from_mgf, load_from_mzml, load_from_msp, load_from_json, and load_from_usi.\n
Boundary markers: The provided documentation does not use delimiters or instructions to treat metadata as untrusted content.\n
Capability inventory: The skill allows for file system operations and network requests to scientific databases.\n
Sanitization: Metadata processing filters like clean_compound_name are intended for data normalization rather than security-focused sanitization.\n- [EXTERNAL_DOWNLOADS]: The skill includes functionality to fetch data from established scientific repositories. The derive_annotation_from_compound_name filter retrieves information from PubChem, and the load_from_usi function fetches spectral data from metabolomics Universal Spectrum Identifier (USI) repositories.

matchms