matchms

Warn

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation and reference files (specifically references/importing_exporting.md) promote the use of load_from_pickle and save_as_pickle. Utilizing the Python pickle module for data deserialization is a well-documented security risk that can lead to arbitrary code execution if a user or agent is induced to load a malicious pickle file.
  • [PROMPT_INJECTION]: There is a discrepancy in the metadata provided in SKILL.md, where the author is listed as 'K-Dense Inc.', which conflicts with the 'qinyan-ai' identity provided in the analysis context. Misleading metadata can be used to bypass safety assessments or misrepresent the skill's origin.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it is designed to ingest and process mass spectrometry metadata from external files (MGF, MSP, JSON, mzML).
  • Ingestion points: Multiple file loading functions documented in references/importing_exporting.md.
  • Boundary markers: The documentation does not specify or encourage the use of delimiters to segregate untrusted spectral metadata from agent instructions.
  • Capability inventory: The skill has the capability to execute Python scripts, perform file system operations, and conduct network requests.
  • Sanitization: While default_filters in references/filtering.md harmonize data for scientific consistency, they do not include mechanisms to sanitize metadata against malicious instructions designed to influence the agent's behavior.
  • [COMMAND_EXECUTION]: The skill instructions in SKILL.md include shell commands for environment setup using uv pip install matchms. While necessary for the skill's operation, it involves execution of commands that should be monitored.
  • [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch data from external scientific repositories, such as retrieving spectra via Universal Spectrum Identifiers (USI) and chemical annotations from PubChem (references/filtering.md). These are recognized as well-known and legitimate services within the metabolomics research community.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 26, 2026, 07:59 PM