The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-supplied images through OCR, creating a surface for indirect prompt injection. Text extracted from images could contain malicious instructions designed to manipulate the agent's behavior.
Ingestion points: Image files processed by Hugging Face API or Tesseract as described in SKILL.md.
Boundary markers: Absent; there are no instructions to the agent to treat OCR output as untrusted or to use delimiters.
Capability inventory: Includes curl for network requests and tesseract and magick for local file processing.
Sanitization: No sanitization, validation, or filtering of the extracted text is implemented.
[EXTERNAL_DOWNLOADS]: The skill connects to an external inference service on Hugging Face to perform OCR tasks.
Evidence: Uses curl to send image data to the Hugging Face Inference API at https://api-inference.huggingface.co/models/qari-ai/QARI-OCR-v0.3.
[COMMAND_EXECUTION]: The skill provides commands for local environment setup and image processing.
Evidence: Includes shell commands for installing tesseract via brew and performing image preprocessing using the magick (ImageMagick) command.

arabic-ocr