The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data (ZIP bundles and Excel workbooks) downloaded from the Jinkō platform, which creates an ingestion point for potentially untrusted content.
Ingestion points: ZIP bundles downloaded from the jinko.ai API and extracted to /tmp/jinko-to-matlab/ as described in scripts/jinko_matlab_parity.py.
Boundary markers: Not explicitly defined in instructions, as the data is processed programmatically rather than interpolated directly into LLM prompts.
Capability inventory: The skill can read/write files and perform network operations via the jinko-sdk as implemented in scripts/jinko_matlab_parity.py.
Sanitization: The skill implements strong security measures, including a safe_extract_zip function that explicitly prevents path traversal (Zip Slip) and symbolic link attacks, and uses yaml.safe_load for configuration files.
[SAFE]: The skill's external dependencies (the jinko Python SDK) and network communications (jinko.ai) are legitimate vendor resources belonging to Nova In Silico.

jinko-task-export-to-matlab