The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/compose.py script invokes the ffmpeg and ffprobe binaries using subprocess.run. While the implementation uses argument lists to prevent direct shell injection, it executes these commands based on inputs provided in the external config.json file.
[EXTERNAL_DOWNLOADS]: The skill requires the pillow Python library for image processing, as documented in the setup instructions within tests/smoke-test.md.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted configuration data from config.json. This data influences the execution of ffmpeg and PIL commands. While the skill employs list-based subprocess calls and specific path escaping for captions to mitigate risks, it lacks comprehensive validation for all configuration-driven paths and processing parameters.

render-absurdist-explainer