The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests scientific model definitions and metadata from the Jinkō platform, which serves as a potential surface for indirect prompt injection if processing untrusted model data.
Ingestion points: Model data is retrieved through client.get_model() in the batch editing, tagging, and validation scripts.
Boundary markers: No specific delimiters or safety instructions are applied to separate model metadata from agent instructions.
Capability inventory: The skill possesses network access via the SDK for API interactions and limited file system access for environment configuration.
Sanitization: The skill relies on the underlying SDK and platform-level validation for sanitizing model components and scientific formulas.
[EXTERNAL_DOWNLOADS]: The skill utilizes external packages and references remote repositories for its core functionality.
Evidence: The skill instructions reference the novainsilico/jinko-skills repository for setup and additional resources.
Evidence: Python scripts import the jinko (SDK) and dotenv modules, which are expected dependencies for this skill's scientific modeling use case.

jinko-model