The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes Python's subprocess.run to execute ffmpeg and yt-dlp commands within its scripts (scripts/cut_aroll.py, scripts/transcribe.py, scripts/qc_static.py). These commands are essential for video processing and resource acquisition. The scripts use the list-based argument format, which mitigates shell injection, but they still operate on user-controlled inputs.
[EXTERNAL_DOWNLOADS]: The skill uses yt-dlp to download video content from external URLs provided by the user. It also references external resources from services such as Unsplash, Pexels, and Hugging Face for media assets and model weights.
[INDIRECT_PROMPT_INJECTION]: The skill transcribes audio from user-provided videos to generate edit lists and scripts. This ingestion of untrusted media content represents a potential surface for indirect prompt injection, where malicious instructions embedded in the audio could attempt to influence the agent's behavior during the editing process.
Ingestion points: User-provided media files processed in scripts/transcribe.py.
Boundary markers: The skill does not implement specific delimiters or 'ignore instructions' warnings to isolate transcribed text from agent instructions.
Capability inventory: Shell execution (ffmpeg), network interaction (yt-dlp, Playwright), and file system operations (symlinking, writing renders).
Sanitization: Transcribed content is not sanitized for potential instruction-like patterns before being processed.

diretor-de-edicao