The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-supplied video transcripts and interpolates the text into prompts for large language models within rewrite_script.py and imagegen_hint.py. This creates a potential surface for indirect prompt injection where adversarial content within a transcript could attempt to influence the agent's behavior during the script restructuring or image planning stages.
Ingestion points: User-controlled transcript text in rewrite_script.py and imagegen_hint.py.
Boundary markers: Not observed in the prompt templates.
Capability inventory: Execution of system commands (FFmpeg, Chrome) and file system access.
Sanitization: None observed on the interpolated text.
[EXTERNAL_DOWNLOADS]: The utils.py script automatically downloads font assets (.ttf, .otf) from official GitHub repositories and well-known content delivery networks (jsDelivr). These downloads are used to provide the necessary typography for video subtitles and covers. The sources (Google Fonts, Adobe, etc.) are established and trusted.
[COMMAND_EXECUTION]: The skill relies on the subprocess module to perform core functionality. This includes executing ffmpeg and ffprobe for media manipulation, and headless Chrome/Chromium in generate_cover_image.py for high-fidelity text rendering. The execution patterns are consistent with the skill's stated purpose.

video-editing