About

A Claude Code skill that tightens long-form recordings — it removes silences, ums, fillers, and dead air while preserving laughs and comedic pauses.

More accurate than cutting from Whisper alone: cuts are driven by Montreal Forced Aligner word boundaries (~10–20ms precision) rather than raw Whisper timestamps (±100–300ms). Whisper produces the transcript text; MFA snaps it to the audio. The result: tighter cuts, no clipped word onsets, and reliable silence detection.

Runs entirely on your machine — no cloud APIs. Tuned against a 7-minute 4K HEVC interview that took 5+ minutes just to decode in other tools; with the right ffmpeg flags the same source cuts in ~30 seconds.

What it does

• Makes a 1080p H.264 proxy with hardware decode so every step runs in seconds
• Transcribes with Whisper, then force-aligns with MFA for the draft cut list
• Tone-aware silence heuristic: aggressive / balanced / sentimental / documentary
• Ground-truths every keep boundary with isolated-window re-transcription
• Renders with trim + concat (not select) so audio never drifts, then verifies the output
• Optional browser timeline editor — canvas waveform, drag-to-delete, live preview

How to use it

1. Run /cut-video and paste a video file path when asked.
2. Pick a tone preset (aggressive is the default), review the proposed cut list, then say go.
3. It renders the cleaned version, verifies the transcript is clean, and opens it for review.