The Agent Skills Directory

[PROMPT_INJECTION]: The skill acts as a data ingestion point by performing OCR on images and providing the text to the agent. This creates a surface for indirect prompt injection if an image contains hidden instructions intended to influence the agent's behavior. * Ingestion points: Image files and macOS clipboard content processed in scripts/cli.js. * Boundary markers: No delimiters or safety warnings are used when returning the extracted text. * Capability inventory: The skill utilizes execSync in scripts/clipboard.js for macOS system integration. * Sanitization: No text filtering or sanitization is performed on the OCR output.
[COMMAND_EXECUTION]: The skill uses execSync in scripts/clipboard.js to run osascript commands for accessing image data on the macOS clipboard. These commands use hardcoded logic and internally managed temporary file paths.
[EXTERNAL_DOWNLOADS]: The tesseract.js library typically downloads language-specific training data from its official remote repositories during initialization, which is the standard and expected operational behavior for this library.

ocr