Video Slicer — 视频智能切片工具
PassAudited by ClawScan on May 16, 2026.
Overview
This appears to be a legitimate local video-slicing skill, but it will install media/AI dependencies, run ffmpeg/Whisper locally, and save frames, audio, and transcripts that may contain private content.
Install in a virtual environment, run the skill only on videos you are allowed to process, use a dedicated private output directory, and review/delete generated frames, audio, transcripts, and clips if the source video is sensitive.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The skill can consume local resources and overwrite generated output files if paths collide.
The skill runs ffmpeg on a user-supplied video path and writes generated frame files. This is central to video slicing, but it is still local command execution with file output and overwrite behavior via ffmpeg's -y option.
subprocess.run(["ffmpeg", "-y", "-ss", ts, "-i", video_path, "-frames:v", "1", "-q:v", "2", out_path], capture_output=True, timeout=30)
Run it only on intended videos, use a dedicated output directory, and review planned paths and time ranges before bulk cutting.
External packages and model files will be installed or downloaded into the user's environment.
The documented setup uses external package installation and Whisper model download/cache. This is disclosed and expected, but versions are not pinned in the user-facing install commands.
pip3 install openai-whisper opencc-python-reimplemented torch torchvision ... 模型下载位置: ~/.cache/whisper/
Use a virtual environment, install from trusted package indexes, consider pinning versions, and keep ffmpeg/Python dependencies updated.
Private speech from the source video may remain on disk as audio and transcript files, and may be included in the agent's working context.
The skill saves extracted audio and transcription outputs locally for later agent analysis. This is purpose-aligned, but those files may contain private spoken content and source-video metadata.
transcript_dir = os.path.join(out_dir, "transcript") ... audio_path = os.path.join(transcript_dir, "segment.aac") ... json_path = os.path.join(transcript_dir, "speech.json") ... txt_path = os.path.join(transcript_dir, "speech.txt")
Choose a private local output folder, avoid synced/shared directories for sensitive videos, and delete transcript/audio artifacts when no longer needed.