The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the tesseract.js package and associated OCR language models at runtime.
Evidence: scripts/image-to-text.sh executes npm install if node_modules is missing.
Evidence: SKILL.md notes that the first run downloads approximately 4MB of language data.
[COMMAND_EXECUTION]: The skill executes a local shell script which in turn invokes a Node.js process to perform the text extraction.
Evidence: SKILL.md instructs the agent to run bash <skill-path>/scripts/image-to-text.sh <image-path> [language].
Evidence: scripts/image-to-text.sh executes node "$SCRIPT_DIR/image-to-text.js" "$@".
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it extracts and returns raw text from untrusted image sources without sanitization or boundary markers.
Ingestion points: The <image-path> argument in scripts/image-to-text.sh allows the agent to process arbitrary image files.
Boundary markers: Absent. The extracted text is returned as a JSON object and presented directly to the agent without delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill executes shell and Node.js scripts via subprocess calls in scripts/image-to-text.sh and scripts/image-to-text.js.
Sanitization: Absent. The extracted text is trimmed but otherwise unmodified before being returned to the agent.

image-to-text