Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+22 more

    Cut Silence

    2

    Automated silence removal, filler word cutting, and -14 LUFS audio normalization for video creators.

    Free

    4 installs5.0 (1 review)
    Cut Silence

    Cut Silence

    Example session with this skill installed

    Remove the silences and filler words from C:/content/lesson_01.mp4. Use a tight margin for the cuts.

    • Read your context and instructions
    • Compiled the cut silence

    Transcribed lesson_01.mp4 (12.2 min, deepgram/nova-3, estimated cost

    $0.05, shown before spending). Nothing has been cut yet: here is the proposal for your approval.

    Proposed cuts

    TypeCountExample
    Fillers (confirmation tags)13"...the account stays active, [ok?] it's the same..."
    Stutters3"today [today] I'm going to show you"
    Restarts (abandoned takes)11see below

    Restarts (full text, what leaves and what stays)

     96.40s  (4.1s)
   OUT:  "Ok, this looks like a lot but it isn't"
   KEEP: "Right, this looks like a lot but it isn't, let me show you"

148.92s  (3.6s)
   OUT:  "the older your profile, the better."
   KEEP: "the older that profile, the better for your structure."

    Kept on purpose: an "Ok," that opens a sentence (connective, not a tag) and one parallel sentence that

    adds information (anaphora, not a restart).

    Settings

    • Base margin 0.1s (tight, as asked) + 61 breathing ranges preserved at sentence ends, 31 of them topic turns
    • Every cut boundary snapped to the waveform valley (up to 150 ms) so no syllable gets clipped

    After you type "apply"

    Original:  12 min 12 s
Final:      9 min 01 s
Saved:      3 min 11 s (26%)
Audio:     -14.1 LUFS · peak -1.0 dBTP  (measured on the delivered file)
Output:    C:/content/edited/lesson_01_edited.mp4  (original untouched)

    Then the edited file is transcribed again to confirm every approved cut actually landed, and any survivor is reported as kept on purpose, rejected earlier, created by the edit or failed to land.

    Next steps

    1. Approve or object to any cut above (just say which one, no JSON editing)
    2. Type apply to render
    3. If it feels rushed, ask for "softer" and the margin goes to 0.3s

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Talking-head videos often suffer from dead air, stutters, and inconsistent volume that kill viewer retention. This skill automates the tedious first pass of editing by surgically removing silence and parasitic speech while ensuring broadcast-standard audio levels.

    What it does

    • Silence removal detects and cuts pauses based on an absolute volume threshold using auto-editor.
    • Audio normalization levels sound to -14 LUFS with a peak limiter to ensure consistency across batches.
    • Filler detection identifies "uhs," "ums," and repetitive stutters through transcription and agent analysis.
    • Duplicate take removal spots where a speaker restarts a sentence after a mistake and cuts the failed attempt.
    • Variable breathing room adjusts margins based on content type, keeping lessons natural and ads tight.

    How it works

    1. Dependency check verifies Python, auto-editor, and ffmpeg are installed and functional in your shell.
    2. Pre-normalization levels the audio so the silence detection threshold remains accurate even for quiet recordings.
    3. Transcription (optional) uses the --fillers flag to map words and pauses for semantic editing.
    4. Approval loop presents proposed filler and duplicate cuts for review before any destructive actions are taken.
    5. Final render executes the cuts and measures the output to verify it hits the -14 LUFS target.

    Frameworks & tools

    Uses auto-editor for timeline manipulation, ffmpeg-normalize and ffmpeg for audio processing, and OpenRouter for optional word-level transcription.

    Why this beats prompting it yourself

    Generic prompts often result in "hallucinated" edits or broken files because they ignore the relationship between recording volume and cut thresholds. This skill enforces a strict "level-then-cut" sequence and uses measured audio metrics rather than assumed targets, preventing files from being accidentally destroyed by aggressive gates.

    Use cases

    • Course creators cleaning up hours of raw lesson footage without losing natural speech pacing.
    • VSL/Ad editors who need speech "glued" together with 0.0s margins for high-energy pacing.
    • YouTubers who need to quickly strip out stutters and failed takes before starting creative editing.

    Known limitations

    Requires an OpenRouter API key for filler word detection. The skill does not perform creative tasks like B-roll overlays, color grading, or captioning. Batch processing is restricted by available RAM.

    How to use

    Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.

    Reviews

    5.0
    1 review
    5
    1
    4
    0
    3
    0
    2
    0
    1
    0

    4 people have installed this skill.

    Trust & safety

    Security scanned

    Verified clean 13 days ago

    • Free to download with an account

    Requires

    Openrouter
    Ffmpeg
    Python
    macOS
    Node.js

    Listed13 days ago

    What's inside

    Frequently Asked Questions