The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation includes several examples of using the pickle format for saving and loading spectral data (load_from_pickle, save_as_pickle). Since Python's pickle module can execute arbitrary code during deserialization, this represents a significant security risk if users load files from untrusted sources.
Evidence: references/importing_exporting.md explicitly lists load_from_pickle and save_as_pickle for fast data handling.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and install the matchms package and its chemistry-related extensions from public registries using the uv tool.
Evidence: SKILL.md contains uv pip install matchms and uv pip install matchms[chemistry].
[COMMAND_EXECUTION]: The skill utilizes shell-level commands for package installation and performs extensive file system operations to read and write mass spectrometry data files in various formats.
Evidence: Code snippets in SKILL.md and references/workflows.md demonstrate reading from and writing to files like output.mgf and processed_data.pkl.
[PROMPT_INJECTION]: The skill ingests untrusted metadata from external spectral library files, creating an attack surface for indirect prompt injection where malicious instructions could be embedded in fields like compound names.
Ingestion points: load_from_mgf, load_from_mzml, and load_from_msp in references/importing_exporting.md.
Boundary markers: Absent; data is parsed into structured objects for similarity calculations without clear isolation for metadata processing.
Capability inventory: The skill has the ability to write files to the local system (save_as_mgf, etc.).
Sanitization: While metadata harmonization filters are applied, they are designed for data consistency rather than filtering for prompt injection patterns.

matchms