The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documentation and examples in SKILL.md and references/importing_exporting.md describe the use of matchms.importing.load_from_pickle and matchms.exporting.save_as_pickle. Because the Python pickle module can execute arbitrary code during deserialization, loading a malformed or malicious pickle file provided by an external source represents a significant security vulnerability.
[INDIRECT_PROMPT_INJECTION]: The skill is built to ingest and process various mass spectrometry file formats (MGF, mzML, MSP, JSON) which can contain arbitrary metadata strings. This represents a potential surface for indirect prompt injection where malicious instructions could be embedded in data files to manipulate the agent's behavior.
Ingestion points: The skill uses functions like load_from_mgf, load_from_mzml, and load_from_json to read external data files into the agent's context.
Boundary markers: The documentation does not specify the use of clear delimiters or instructions for the agent to ignore potentially malicious content within these data files.
Capability inventory: The skill has capabilities to read from and write to the file system (save_as_mgf, save_as_json, save_as_pickle) and perform complex mathematical similarity calculations.
Sanitization: The provided 'default_filters' perform metadata harmonization and cleaning for analytical purposes but do not include security-focused sanitization to prevent prompt injection attacks.
[DATA_EXFILTRATION]: The derive_annotation_from_compound_name function mentioned in references/filtering.md performs automated lookups to the PubChem database to retrieve chemical structures. While this is a standard feature for metabolomics research, it involves making outbound network requests to an external service.

matchms