The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill creates a significant attack surface by converting visual data (images) into natural language text which is then processed by the agent.
Ingestion points: scripts/ocr.py (image_file) and scripts/ocr_url.py (image_url).
Boundary markers: Absent. The extracted text is written directly to the output file without any delimiters or 'ignore instructions' headers.
Capability inventory: The skill writes to local files (open(output_path, 'w')) and performs network requests via requests.get.
Sanitization: None. The script does not validate or sanitize the OCR output before saving it for the agent's consumption.
[External Downloads] (MEDIUM): scripts/ocr_url.py uses the requests library to download images from arbitrary, untrusted URLs provided at runtime.
[Privilege Escalation] (LOW/INFO): The SKILL.md file instructs users to use sudo apt-get install or sudo yum install for system dependencies. While standard for installation, it highlights the requirement for elevated privileges to set up the environment.

ocr