The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/transcribe_pages.py invokes external binaries (libreoffice, pdftoppm, and tesseract) to handle document conversion and OCR. These commands are executed using argument lists rather than shell strings, which effectively prevents command injection risks from malicious file names.\n- [DYNAMIC_EXECUTION]: The script dynamically modifies the Python sys.path to import modules from adjacent skill directories such as /mnt/skills/user/browsing-bluesky/scripts. This facilitates code reuse across the platform and targets hardcoded internal paths.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user documents which serves as an entry point for indirect instructions embedded in the processed images.\n
Ingestion points: scripts/transcribe_pages.py (via PPTX, PDF, and image file parsing).\n
Boundary markers: The vision models are guided by a specific transcription prompt (_PAGE_PROMPT) that defines the expected task and output structure.\n
Capability inventory: File system access for reading/writing images and subprocess execution for conversion tools.\n
Sanitization: The skill does not perform sanitization of the text content extracted from processed pages before passing it to the agent.

transcribing-images