The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically installs missing dependencies including yt-dlp and mlx-whisper using pip from a well-known mirror (Tsinghua University) and ffmpeg via Homebrew on macOS.
[COMMAND_EXECUTION]: Utilizes subprocess calls to execute yt-dlp for media retrieval, ffmpeg for audio extraction, and system utilities (like open) to display the generated HTML report.
[DATA_EXFILTRATION]: Manages Netscape-format cookie files for authenticated downloads from media sites. These are stored and read locally from ~/Downloads/fetch-what-say/cookies/ and are not transmitted to third-party servers.
[PROMPT_INJECTION]: As a tool that processes external media, it possesses a surface for indirect prompt injection (Category 8) if a processed video transcript contains instructions intended to manipulate the agent's summarization output.
Ingestion points: transcript.txt generated from external URLs or local files in scripts/fetch_what_say.py.
Boundary markers: The summary prompt defines a strict structure but does not use specific delimiters to isolate the transcript content from instructions.
Capability inventory: The skill can write files to the local system and execute media processing commands.
Sanitization: No sanitization or filtering is performed on the transcript text before it is provided to the agent for summarization.

fetch-what-say