The Agent Skills Directory

[DYNAMIC_EXECUTION]: The file test_snippets.py uses the exec() function to dynamically execute Python code blocks extracted from SKILL.md. This pattern allows for the execution of arbitrary code contained within the markdown file, which poses a risk if the file's contents were to be maliciously altered.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to fetch and process data from a wide variety of external biomedical databases and tools, such as DisGeNet, MSigDB, and UniProt. This architecture is vulnerable to indirect prompt injection, where an attacker who can influence the content of these databases might inject instructions to manipulate the agent's behavior.
Ingestion points: API responses and database lookup results from the various biomedical sources listed in the SKILL.md routing table.
Boundary markers: There are no explicit instructions or delimiters provided to ensure the agent isolates and treats external tool output as untrusted data.
Capability inventory: The agent has access to a Python code interpreter and network-enabled tools to fulfill the queries described in the skill.
Sanitization: The instructions do not specify any validation or sanitization of the data retrieved from external tools before it is used in computations or reasoning.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md explicitly direct the agent to generate and execute Python code snippets for tasks like ORF counting and restriction digest calculations. While this is the intended utility, the execution of generated code is a high-privilege action that carries inherent security risks.

tooluniverse-biomedical-fact-lookup