The Agent Skills Directory

[SAFE]: The skill implementation follows secure coding practices. Commands are structured as plans for explicit execution, and file paths are resolved using standard path utilities. No persistence, privilege escalation, or data exfiltration mechanisms were detected.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external PDF files, which represents a potential attack surface if those files contain adversarial instructions.
Ingestion points: Files are read and parsed via extract-text.ts, extract-tables.ts, and ocr-pdf.ts using external libraries.
Boundary markers: The instructions do not define specific delimiters to separate untrusted document content from agent instructions, though the output is structured (JSON/CSV).
Capability inventory: The scripts allow for local file system operations (read/write) related to PDF and image processing.
Sanitization: create-pdf.ts contains an internal character mapping for PDF encoding compatibility (sanitizeForPdf), and extract-tables.ts performs standard CSV escaping for extracted data.
[EXTERNAL_DOWNLOADS]: The skill references a well-known library that may perform runtime downloads of necessary data files.
Evidence: The ocr-pdf.ts script utilizes tesseract.js to initialize an OCR worker (Tesseract.createWorker(args.lang)), which typically fetches language-specific trained data models from its official CDN when requested.

pdf-toolkit