drugbank-database

Warn

Audited by Gen Agent Trust Hub on Mar 31, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill uses the pickle module in references/data-access.md to cache and load parsed database results. Deserializing data with pickle.load() can allow for arbitrary code execution if the cache file is modified by a malicious actor.\n- [DATA_EXFILTRATION]: The skill performs network requests to go.drugbank.com to download database updates and query the REST API, as documented in references/data-access.md. These operations target the official service provider for the database.\n- [EXTERNAL_DOWNLOADS]: The skill specifies several Python packages for installation in SKILL.md, including drugbank-downloader, rdkit, and bioversions. these are standard, well-known libraries in the bioinformatics and data science communities.\n- [PROMPT_INJECTION]: The skill downloads and processes external XML database files, which presents a surface for indirect prompt injection.\n
  • Ingestion points: XML parsing logic located in references/drug-queries.md and scripts/drugbank_helper.py.\n
  • Boundary markers: There are no explicit boundary markers or instructions to ignore content within the ingested XML data.\n
  • Capability inventory: The skill has capabilities for file system writes (via pandas CSV export) and network access (via requests). \n
  • Sanitization: No input validation or sanitization is performed on the content extracted from the XML database entries.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 31, 2026, 08:21 AM
Security Audit — agent-trust-hub — drugbank-database