The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The example code in references/patterns.md defines a calculator tool that directly invokes Python's eval() on a user-influenced string. This allows an attacker to execute arbitrary Python code by providing strings containing system imports or dangerous built-ins.
Evidence (references/patterns.md): def calculator(expression: str) -> str: """Evaluate a math expression.""" return str(eval(expression))
[DYNAMIC_EXECUTION]: The use of dynamic string evaluation without any sanitization or sandboxing is a high-risk pattern. In the context of an AI agent, this can be exploited via both direct and indirect prompt injection to perform operations outside the intended scope of a math utility.
[COMMAND_EXECUTION]: Through the eval() function, the skill's reference code provides a pathway for executing shell commands (e.g., using __import__('os').system()), which could lead to environment compromise or data theft.

langgraph