The Agent Skills Directory

[DYNAMIC_EXECUTION]: The calculator tool implementation in the 'Basic Agent Graph' section of SKILL.md uses the eval() function to process the expression input string. Using eval() on untrusted input allows for arbitrary Python code execution within the environment where the agent is running. Documentation should recommend safer alternatives like ast.literal_eval or a dedicated math expression parser.
[INDIRECT_PROMPT_INJECTION]: The skill describes an agent architecture that processes external user input and routes it to tools with significant capabilities.
Ingestion points: Untrusted user input enters the system through the messages list in AgentState within SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the prompt templates or node definitions.
Capability inventory: The system includes a calculator tool utilizing eval(), which can be exploited to execute arbitrary code if the LLM is manipulated by input data.
Sanitization: The implementation lacks any input validation, escaping, or filtering before passing data to the eval()-based tool.

langgraph