tooluniverse-protein-design-campaign
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a local Python script,
scripts/preflight_submission.py, to validate candidate CSVs and methods files. The script performs read-only operations using standard Python libraries (argparse, csv, hashlib, re) to ensure data integrity and detect accidental information disclosure. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external CSV files (
candidates.csv) and text files (methods.txt), which presents an indirect prompt injection attack surface. To mitigate this, the included preflight script explicitly scans for and flags common injection patterns, such as 'ignore previous instructions' or 'system prompt', before the data is processed by the agent. - [DATA_EXFILTRATION]: The skill incorporates active data-leak prevention through its validation script. The utility uses regular expressions to scan inputs for sensitive patterns, including hardcoded API keys, private cryptographic keys, and internal network addresses, ensuring that these are not included in final submission packages.
Audit Metadata