The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run to execute external binaries including ffmpeg, ffprobe, and yt-dlp. These are fundamental to the skill's primary purpose of video editing and rendering.
[DYNAMIC_EXECUTION]: The file scripts/transcribe.py uses importlib.util to dynamically load and execute a Python module from a relative file path (../video-understand/scripts/transcribe.py). While this facilitates code reuse between sibling skills, it introduces a dependency on the integrity of the external skill directory.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it processes untrusted data from external sources.
Ingestion points: External media downloaded via yt-dlp from user-provided URLs and the resulting transcripts generated by faster-whisper (referenced in scripts/analyze.py and scripts/verify_cut.py).
Boundary markers: There are no explicit instructions or delimiters used when the agent reads the transcript to ignore or sanitize embedded instructions.
Capability inventory: The skill possesses powerful capabilities through shell execution of ffmpeg and yt-dlp (found in scripts/cut_render.py and scripts/selfcheck_frames.py).
Sanitization: No specific sanitization or validation of the transcript text or media metadata is implemented before the content is processed by the agent to make editorial decisions.

video-cut