The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/roadshow_composer.py executes system commands via subprocess.run to invoke FFmpeg. These commands are used to create slideshows from images, mix audio tracks, and burn subtitles into the final video output. While the implementation uses the safer list-based argument format to prevent shell injection, the execution of system tools on data derived from user input remains a sensitive capability.
[EXTERNAL_DOWNLOADS]: The skill documentation in SKILL.md requires several external Python dependencies, including moviepy, pillow, pydub, and requests. It also instructs users to install the ffmpeg system utility through package managers like apt-get or brew to enable video processing functionality.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it ingests and processes untrusted data from user documents and brand examples to drive its multi-role generation workflow.
Ingestion points: User-provided text documents are analyzed in 'Role One' and design materials (PPTs, images) are analyzed in 'Role Three' to extract content and style guidelines.
Boundary markers: There are no explicit delimiters or instructions provided to the agent roles to isolate user-provided content from the agent's task-oriented instructions.
Capability inventory: The skill possesses capabilities for file system writes, network requests (via TTS API), and system command execution (ffmpeg).
Sanitization: The skill lacks explicit sanitization of user-derived text before it is interpolated into the generated roadshow scripts or utilized in the parameters for video synthesis filters.

ppt-roadshow-generator