The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill exposes the load_from_pickle function (documented in references/importing_exporting.md), which uses Python's native pickle module for data deserialization. Loading data from untrusted pickle files is a known security risk that can lead to arbitrary code execution in the agent's environment.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve data from external services:
load_from_usi fetches spectral data from online repositories via Universal Spectrum Identifiers.
derive_annotation_from_compound_name (in references/filtering.md) retrieves chemical annotations from PubChem, a well-known scientific service provided by the National Institutes of Health (NIH).
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection due to its ingestion of untrusted data from multiple file formats and remote services.
Ingestion points: Data enters the agent's context through load_from_mgf, load_from_mzml, load_from_json, load_from_usi, and external metadata enrichment from PubChem.
Boundary markers: The documentation does not specify the use of clear delimiters or instructions to ignore embedded commands within the processed spectral data.
Capability inventory: The skill can execute file system writes (save_as_mgf, save_as_pickle), complex data transformations, and network requests to external databases.
Sanitization: While the skill provides extensive metadata harmonization and peak filtering functions, these are designed for scientific data standardization rather than security-focused sanitization of potentially malicious payloads.

matchms