The Agent Skills Directory

[SAFE]: The skill provides documentation and examples for standard OpenAI-compatible tool-calling loops. All network communication is directed to the vendor's own API endpoint at api.runware.ai.
[SAFE]: The implementation guidelines include critical security best practices, such as bounding tool-calling loops to prevent infinite execution, validating and parsing JSON arguments safely, and returning errors as data rather than allowing exceptions to crash the runtime.
[SAFE]: The skill emphasizes a 'least privilege' design for tools, instructing users to constrain arguments with schemas and enums, and noting that the model does not have direct access to internal systems, only to the JSON results returned by the user's own functions.

llm-agent