The Agent Skills Directory

[DYNAMIC_EXECUTION]: The test utility test_snippets.py uses the Python exec() function to execute code blocks parsed from the SKILL.md file. While intended for validation of documentation snippets, this provides an internal attack surface where modifying the markdown file leads to arbitrary code execution.
[COMMAND_EXECUTION]: The skill provides explicit instructions and Python templates for the agent to generate and run code locally. This is used for deterministic computations like Open Reading Frame (ORF) counting, restriction enzyme digestion modeling, and Mendelian genetics ratios. This creates a surface for dynamic code generation and execution based on agent reasoning.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from various external biomedical databases (e.g., DisGeNet, UniProt, ClinVar) and user-provided questions. While these are reputable scientific sources, the skill lacks explicit boundary markers or sanitization logic when interpolating external tool outputs back into the agent's context, making it potentially susceptible to instructions hidden within database records.
[DATA_EXPOSURE]: The skill instructions reference the use of sensitive identifiers such as DISGENET_API_KEY and OMIM keys. While it correctly advises that these keys are required for certain lookups, there is a secondary risk if the agent is induced to reveal these environment-stored secrets through adversarial questioning.

tooluniverse-biomedical-fact-lookup