The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting word text from external JSON files and embedding it into generated subtitle artifacts.\n
Ingestion points: The words_json input file in scripts/make_subtitles.py is the primary source of untrusted text.\n
Boundary markers: The generated ASS subtitle files do not include explicit delimiters or instructions to distinguish the injected text from control content for downstream processors.\n
Capability inventory: The skill possesses file system write capabilities through Python's pathlib and PIL libraries across all scripts for generating subtitles and PNG overlays.\n
Sanitization: The scripts/make_subtitles.py script performs basic escaping by replacing curly braces with parentheses to maintain the integrity of the subtitle format, but it does not perform semantic validation of the content.\n- [EXTERNAL_DOWNLOADS]: The skill documentation and scripts specify a dependency on the requests library to fetch graphical assets (emojis) from a Content Delivery Network (CDN) and the Pillow library for image processing.

style-captions