alterlab-drugbank
Warn
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The code examples in 'references/data-access.md' demonstrate the use of the Python 'pickle' module to load cached results from a local file ('drugbank_parsed.pkl'). Because 'pickle.load()' is capable of executing arbitrary code embedded in the serialized data, this pattern is insecure if the cache file can be manipulated by an untrusted source.
- [EXTERNAL_DOWNLOADS]: The skill facilitates the download of the DrugBank database from 'go.drugbank.com' using the 'drugbank-downloader' package. This represents a dependency on external datasets and infrastructure.
- [CREDENTIALS_UNSAFE]: The skill provides instructions for storing DrugBank authentication credentials in environment variables or configuration files, specifically mentioning '~/.config/drugbank.ini'. This practice exposes sensitive secrets to the agent's environment and the local file system.
- [PROMPT_INJECTION]: The skill ingests and processes large amounts of external XML data from DrugBank. The instructions lack explicit boundary markers or sanitization guidelines to prevent the agent from potentially interpreting embedded text within the database as instructions, creating a surface for indirect prompt injection.
Audit Metadata