The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/jump_cut_vad_singlepass.py utilizes torch.hub.load to download and execute code from the snakers4/silero-vad repository on GitHub. This constitutes execution of remote code from a source that is not on the predefined trusted list.
[COMMAND_EXECUTION]: All scripts in the toolkit make extensive use of shell command execution via subprocess.run to call external utilities including ffmpeg, ffprobe, yt-dlp, and npx. This is found in scripts/auto_captions.py, scripts/compress_video.py, scripts/insert_3d_transition.py, scripts/jump_cut_vad_singlepass.py, scripts/simple_video_edit.py, scripts/vertical_crop.py, and scripts/youtube_clip.py.
[EXTERNAL_DOWNLOADS]: The skill performs external downloads from various sources. scripts/youtube_clip.py downloads video files using yt-dlp, and scripts/jump_cut_vad_singlepass.py downloads the Silero VAD model and code from GitHub.
[DATA_EXFILTRATION]: The skill transmits data to external services. scripts/simple_video_edit.py and scripts/youtube_clip.py send transcripts and video metadata to anthropic.com for processing, and scripts/simple_video_edit.py uploads video files to auphonic.com. These operations are consistent with the skill's stated purpose but involve the transfer of user data to third-party platforms.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection in scripts/youtube_clip.py and scripts/simple_video_edit.py, where external content is processed by an LLM.
Ingestion points: Transcripts generated from external YouTube videos or user-provided video files are used to populate prompts for Claude.
Boundary markers: The prompts in scripts/youtube_clip.py and scripts/simple_video_edit.py do not implement delimiters or instructions to ignore commands embedded within the transcript text.
Capability inventory: The skill has high local capabilities, including arbitrary command execution via FFmpeg and network access for data uploads.
Sanitization: There is no evidence of sanitization or filtering of the transcript content before it is processed by the AI model.

video-edit