The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/transcribe.py and scripts/transcribe_windows.py utilize subprocess.run and subprocess.check_call to execute shell commands, including pip install, apt-get install, yt-dlp, and ffmpeg.
[EXTERNAL_DOWNLOADS]: The skill performs automated package installations from PyPI and fetches media content from external URLs provided by the user.
[REMOTE_CODE_EXECUTION]: The function ensure_pip_package in scripts/transcribe.py employs __import__ for dynamic module loading at runtime.
[DATA_EXFILTRATION]: The skill accesses the OPENAI_API_KEY from the environment and transmits audio data to external transcription service endpoints.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: External video metadata (titles, uploader names) and transcript text are ingested into the agent's context from various platforms in scripts/transcribe.py and scripts/transcribe_windows.py.
Boundary markers: The skill lacks explicit 'ignore embedded instructions' markers or delimiters when presenting ingested metadata to the agent.
Capability inventory: The skill possesses capabilities for file writing, network access, and system command execution.
Sanitization: A slugify function is used to sanitize filenames derived from external metadata to prevent path traversal or shell injection.

video-transcriber