The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/validate_skill.py downloads a sample image from the official Tesseract documentation repository on GitHub (raw.githubusercontent.com/tesseract-ocr/tessdoc) using urllib.request. This is a legitimate diagnostic step from a well-known source.
[COMMAND_EXECUTION]: The skill executes the Tesseract binary as a subprocess via the pytesseract library. The command path is configurable in the ocr_config.json files, which is a standard requirement for Tesseract wrappers.
[PROMPT_INJECTION]: The skill implements a regex-based parser to extract file paths from a natural language request string. While this involves processing user input, it is scoped to file path identification for OCR. The skill also presents an indirect prompt injection surface (Category 8) because it extracts text from external images which could contain malicious instructions. Mandatory Evidence Chain: 1. Ingestion points: request field in ocr_config.json and processed image content; 2. Boundary markers: Absent; 3. Capability inventory: Subprocess execution and local file access; 4. Sanitization: Regex for path extraction, but no sanitization for OCR output.

image-ocr