The Agent Skills Directory

[COMMAND_EXECUTION]: The calculator tool in examples/02_react_agent.py utilizes the Python eval() function to process user-supplied mathematical expressions. Because the input to this function is derived from LLM responses that process untrusted user messages, an attacker can craft a prompt to execute arbitrary code on the host system.
[PROMPT_INJECTION]: The skill contains an attack surface for indirect prompt injection by combining untrusted data ingestion with dangerous tool capabilities. (1) Ingestion points: The messages list in examples/02_react_agent.py receives raw user input. (2) Boundary markers: There are no delimiters or instructions provided to separate user-controlled data from system instructions or tool execution triggers. (3) Capability inventory: The agent has access to the eval()-based calculator tool in the same example file. (4) Sanitization: The skill does not perform any validation, escaping, or filtering of the expression string before passing it to the eval() function.

langgraph-agent