Work With Media

Turn audio and video into text on macOS. Two complementary skills:

• macwhisper-cli — wraps the MacWhisper mw CLI for local-file transcription. whisper-cpp, WhisperKit, Parakeet, and Apple’s on-device Speech framework — all running locally.
• yt-dlp — pulls captions, descriptions, chapters, and audio from YouTube, Vimeo, and hundreds of other sites. Leads with a subs-first strategy (no audio download) so a TLDR doesn’t require a full transcription. Can also output transposed (pitch-shifted) MP3s from a URL via ffmpeg.

The skills compose: for a YouTube URL, the yt-dlp skill usually finishes the job from captions alone; when a video has no captions or the user needs high accuracy, yt-dlp downloads the audio and hands off to macwhisper-cli for transcription.

Installation

Run these in a terminal (not inside a Claude Code session):

claude plugin marketplace add jtsternberg/claude-plugins
claude plugin install work-with-media@jtsternberg

Prerequisites

macwhisper-cli skill

• MacWhisper (macOS only) — installed with the CLI enabled (Settings → Advanced → Install under Command-Line Tool). This drops /usr/local/bin/mw.
• At least one transcription model downloaded from MacWhisper’s UI.

yt-dlp skill

• yt-dlp on PATH. Required for every URL workflow.
• ffmpeg on PATH. Required only for the audio-transcription fallback and transposed-audio paths.

brew install yt-dlp ffmpeg

Non-Homebrew install options: yt-dlp · ffmpeg.

Why MacWhisper?

MacWhisper is the speech-to-text tool I reach for on macOS. It’s positioned as a file-transcription app, but ships a first-class system-wide dictation mode and a full model catalog in the same package — for a single one-time payment, with everything running on-device.

• One-time purchase, no subscription. The speech-to-text space on Mac is crowded — Superwhisper, Wispr Flow, Voibe, Speakmac, Aiko, WhisperClip, and others — and most are billed yearly or monthly. MacWhisper is pay-once-own-forever.
• File transcription and dictation in one app. Drag in audio/video for transcription (what this plugin wraps), or hit a hotkey anywhere in macOS to dictate straight into the active text field. Superwhisper, Wispr Flow, and Voibe are primarily dictation tools — file transcription is a separate problem they don’t really solve. MacWhisper handles both workflows natively.
• On-device, using the best open models. Whisper (whisper-cpp, WhisperKit), Parakeet, and Apple’s Speech framework (macOS 26+) all run locally — audio never leaves your machine. No cloud upload, no per-minute API bill, no data-retention worries. You pick the accuracy/speed tradeoff that fits each job.

If you don’t own MacWhisper yet: https://goodsnooze.gumroad.com/l/macwhisper. The macwhisper-cli skill wraps MacWhisper’s built-in mw CLI, so the app itself is required.

See more about MacWhisper in this video.

Note: I am not affiliated, just a fan.

Structure

plugins/work-with-media/
├── .claude-plugin/plugin.json
├── README.md
├── shared/
│   ├── output-conventions.md     ← how both skills manage /tmp files + Read-on-demand
│   └── routing.md                ← decision matrix for which skill handles which input
└── skills/
    ├── macwhisper-cli/           ← local-file transcription
    │   ├── SKILL.md
    │   └── references/
    │       └── choosing-a-model.md
    └── yt-dlp/                   ← URL text/audio extraction
        ├── SKILL.md
        └── references/
            └── setup.md

Both skills link to the files under shared/ rather than inlining the conventions, so the two skills can’t drift apart on temp-file naming, Read-on-demand defaults, or the subs-first-vs-transcription decision.

Additional Documentation

• skills/macwhisper-cli/SKILL.md — MacWhisper CLI wrapper: live CLI reference, local-file workflows, troubleshooting.
• skills/macwhisper-cli/references/choosing-a-model.md — engine-by-engine guidance and how to change the default model.
• skills/yt-dlp/SKILL.md — yt-dlp workflows: subs-first TLDRs, audio-transcription fallback, direct-URL handling.
• skills/yt-dlp/references/setup.md — install yt-dlp + ffmpeg.
• shared/output-conventions.md — /tmp/ naming + Read-on-demand defaults for both skills.
• shared/routing.md — which skill handles which input.
• MacWhisper CLI docs — upstream reference.