The Agent Skills Directory

Dynamic Code Evaluation: The calculate tool examples in SKILL.md (both Python and TypeScript) utilize the eval() function to process mathematical expressions. While functional for demonstration, this pattern allows for the execution of arbitrary code if the input string is not strictly validated or contains unintended commands. It is recommended to use safer alternatives such as ast.literal_eval in Python or a dedicated math expression parser.- Indirect Prompt Injection Surface: The skill defines tools such as search and calculate that process inputs which may originate from external sources or user-provided data. The examples do not demonstrate the use of boundary markers or sanitization, which creates a potential surface where instructions embedded in processed data could influence the agent's behavior. Using structured schemas and explicit instructions to ignore embedded commands can help mitigate this area for review.

langchain-fundamentals