Trim filler

Skill jperrello/C0BALT_CUT/.claude/skills/trim-filler

Investigation on using Claude Code to automatically generate profitable YouTube videos.

Install

npx -y skills add jperrello/C0BALT_CUT --skill trim-filler

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Semantic dead-air / filler removal. Claude reads a clip-local word-timed transcript and marks filler words, trail-offs, false starts, repeated re-starts, and short digressive asides for removal. Emits keeps.json (ranges to keep) and transcript.trimmed.json (kept words with shifted timestamps). Pairs with cut-filler, which applies the cuts to the clip's video.

SKILL.md

2.4 KB, 561 tokens by cl100k_base, as published. Nobody here has run it

trim-filler

Tightens a podcast clip by stripping low-value words, not just silence. Where tighten-pace cuts inter-word gaps > gap_max, trim-filler cuts whole spans of speech that don't carry the point — "uh", "um", "you know what I mean", false starts, and the brief tangents speakers fall into mid-sentence.

Example: "I opened a restaurant that was like uhm — I love restaurants haha — yeah me too anyways it sold burgers" → "I opened a restaurant that sold burgers".

Invoke

.claude/skills/trim-filler/trim-filler.sh <in_transcript> <out_keeps> <out_transcript> [pad=0.05]

in_transcript — clip-local word-timed transcript JSON (output of transcribe rebased to the clip)
out_keeps — written keeps.json containing the keep ranges and removal rationale
out_transcript — written clip-local transcript JSON with only kept words and shifted timestamps
pad — kept padding around each retained span (sec)

Output shape — keeps.json

{
  "source": "<original transcript path>",
  "keeps": [[0.0, 4.21], [5.83, 12.04], ...],
  "removed": [
    {"t0": 4.21, "t1": 5.83, "words": "uhm — I love restaurants haha — yeah me too anyways", "reason": "filler + digression"}
  ],
  "removed_total": 1.62
}

How

Build a numbered transcript (one word per line: <idx>\t<t0>\t<t1>\t<word>).
Ask Claude (claude -p) which index ranges to REMOVE because they are filler, trail-offs, false starts, or short asides that don't carry the speaker's point. Speech that delivers the actual content stays.
Parse Claude's JSON reply, build the complement (keeps), pad each kept span, merge overlaps, write both outputs.

Re-runs are idempotent via an mtime+pad signature in <out_keeps>.tfmeta.

Pairs with

cut-filler — consumes keeps.json and re-encodes the clip's video+audio.

Caveats

Operates per-clip after cut-clip + rebase. Don't run on the full source transcript — Claude context budget.
If Claude returns "no cuts", outputs are pass-through.