The Agent Skills Directory

[SAFE]: The skill's instructions and associated Python scripts are focused on legitimate operations within the Jinkō platform ecosystem. No malicious patterns such as prompt injection, data exfiltration, or persistence mechanisms were detected.
[INDIRECT_PROMPT_INJECTION]: Several scripts provide the capability to ingest data from local JSON files, creating an attack surface for indirect prompt injection if those files were to contain malicious instructions.
Ingestion points: The --from-json argument in scripts/create_advanced_output_set.py and the --add-objective-json argument in scripts/edit_advanced_output_set.py read content from specified file paths.
Boundary markers: The data is ingested as structured JSON, which provides inherent separation from executable logic.
Capability inventory: Ingested data is used as parameters for Jinko SDK calls to create or modify scoring designs.
Sanitization: The load_components_from_json and load_objective_json functions perform comprehensive validation of the JSON schema, including type checking for every field and verifying numeric constraints (e.g., wide/narrow range ordering and positive weight values) before any SDK operations are performed.
[EXTERNAL_DOWNLOADS]: The skill references and requires the jinko-sdk library and the jinko-sdk-setup skill. These are vendor-owned resources from Nova In Silico designed for secure integration with their platform.

jinko-output-set