tooluniverse-binder-discovery
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface due to its reliance on external, third-party data sources. Ingestion points: The agent fetches information from ChEMBL, PubChem, PubMed, BioRxiv, and MedRxiv. Boundary markers: The instructions lack explicit markers or delimiters to segregate this external content or warn the agent to ignore potential instructions embedded within scientific abstracts or data. Capability inventory: The agent has the capability to write to the local filesystem (creating .md, .csv, and .json files) and execute further tool calls based on the findings. Sanitization: There is no documented process for sanitizing or validating the incoming scientific data.
- [COMMAND_EXECUTION]: The skill includes instructions for the agent to use Python snippets for environment verification, such as checking for the existence of an 'NVIDIA_API_KEY' using 'os.environ.get'. These operations interact directly with the agent's runtime environment variables.
- [PROMPT_INJECTION]: The instructions mandate a 'Report-First' workflow that explicitly directs the agent to suppress tool outputs and the intermediate search process from the user interface. While intended to provide a cleaner user experience, this behavior reduces the transparency of the agent's autonomous tool usage.
- [SAFE]: The skill's integration with reputable scientific platforms like RCSB PDB, UniProt, and NVIDIA NIM for structure prediction and docking validation is appropriate for its stated purpose. The use of environment variables for API keys follows standard security practices for managing tool credentials.
Audit Metadata