The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from PDF files and bibliographic metadata, which could potentially contain malicious instructions intended to influence the agent. Ingestion points: Uses create_reference_from_pdf and create_extract_from_pdf_quote to ingest external document content (SKILL.md, references/sdk-workflow.md). Boundary markers: Includes a 'Core Rule' in SKILL.md stating to treat PDFs and bibliography metadata as project data, never as instructions. Capability inventory: File manipulation and data extraction methods via the jinko-sdk, including save_content_to_file and create_extract_from_pdf_quote (references/sdk-workflow.md). Sanitization: No specific sanitization or filtering of the PDF text layer is detailed beyond the SDK's internal extraction logic.
[EXTERNAL_DOWNLOADS]: The skill references the installation of the jinko-sdk and related tools. Evidence: Mentions requirements for jinko-sdk[pdf] and suggests installation of setup skills from the novainsilico/jinko-skills repository, which are identified as vendor resources (SKILL.md).

jinko-reference