The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads YouTube transcripts, storyboard images, and video audio streams using the yt-dlp utility. It also references and clones the official whisper.cpp repository from GitHub for local transcription. These downloads are necessary for the skill's core functionality and target well-known services.
[COMMAND_EXECUTION]: Multiple Python scripts (e.g., whisper_dispatcher.py, ytdlp.py) use subprocess.run to execute system binaries including ffmpeg, yt-dlp, and whisper-cli. The implementation follows security best practices by passing arguments as lists and using shutil.which for path resolution, minimizing the risk of shell injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes live stream transcripts and chat logs which are external, untrusted data sources. While this presents an injection surface, the skill includes specialized validation logic (e.g., corruption.py, validate_part_coherence.py) to detect and handle ASR errors, repetition loops, and topic hallucinations.
[DYNAMIC_EXECUTION]: The script markdown.py uses __import__('sys') for a one-liner print to stderr, and pack_timestamps.py uses importlib.util to load a sibling script. Both instances use hardcoded, local targets and do not execute untrusted code.
[DATA_EXFILTRATION]: The skill includes instructions for synchronizing generated dictionaries and timestamp archives with the author's GitHub repositories. This is a documented archival workflow and does not involve sensitive system or credential files.

anibon-timestamper