drugbank-database

Warn

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: Use of the pickle module for caching parsed database content in references/data-access.md. Deserializing untrusted data with pickle.load() is unsafe because it can lead to arbitrary code execution. An attacker who gains access to the local filesystem could replace the drugbank_parsed.pkl file to execute malicious code when the agent loads the cache.
  • [EXTERNAL_DOWNLOADS]: The skill downloads the DrugBank database from go.drugbank.com using the drugbank-downloader package. This is the intended behavior of the skill and targets an established bioinformatics service.
  • [COMMAND_EXECUTION]: Instructions provided in SKILL.md and references/data-access.md guide the user to execute shell commands for package installation and database management using pip and the drugbank_downloader CLI.
  • [PROMPT_INJECTION]: The skill processes external XML data from DrugBank (references/drug-queries.md). While there are no explicit boundary markers to prevent indirect prompt injection from malicious drug descriptions, the risk is limited as the data originates from a curated scientific database and the skill does not use high-privilege capabilities to process the text.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 26, 2026, 07:58 PM