The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides a code example for a calculator tool (in SKILL.md) that uses the eval() function to process input strings. Using eval() allows for the execution of arbitrary Python code, which presents a significant security risk if the input (the expression argument) is derived from untrusted sources, such as model-generated content or user prompts.
[INDIRECT_PROMPT_INJECTION]: The skill defines a framework for processing external data without sufficient security controls.
Ingestion points: The app.invoke() method in the 'Basic Agent Graph' example (in SKILL.md) accepts a messages list which typically contains untrusted user input.
Boundary markers: There are no delimiters or instructions provided to the agent to treat the messages content as data rather than instructions.
Capability inventory: The defined nodes include a search tool (network/information access) and a calculator tool (eval() based code execution).
Sanitization: The skill lacks any evidence of input validation, output filtering, or sanitization logic to prevent embedded instructions in user messages from hijacking the agent's control flow.

langgraph