agentsclimarketplace

Bookend trim

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

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

Install
npx -y skills add jperrello/C0BALT_CUT --skill bookend-trim

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

Snap each picked segment's [t0, t1] to a sentence boundary so shorts begin and end on full thoughts instead of mid-sentence. Claude reads ±extend seconds of context per span and picks clean sentence-completing endpoints from whisper transcript-line boundaries. Runs after verify-coherence, before cut-clip.

SKILL.md

2.0 KB, as published. Nobody here has run it

bookend-trim

Picked segments tend to begin or end mid-sentence because pick-segments and verify-coherence choose by topic, not prosody. This skill nudges each span's [t0, t1] to the nearest clean sentence boundary so the resulting short feels like a complete thought.

Invoke

.claude/skills/bookend-trim/bookend-trim.sh <segments.json> <transcript.json> <out.json> [extend=6.0] [dmin=20]
  • segments: output of verify-coherence (or pick-segments)
  • transcript: full-source transcript (whisper output with segments[])
  • out: same shape as input, with t0 / t1 adjusted and a bookend_note
  • extend: max seconds Claude may push outward in either direction (default 6.0)
  • dmin: min surviving duration; an adjustment that collapses below this is rejected

How

This project's transcribe skill runs whisper.cpp with --max-len 1 --split-on-word, which strips punctuation. So a pure ends-with-./?/! snap finds nothing and inter-word gaps are also collapsed. Instead we delegate to Claude:

  1. For each span, build a context window: every whisper transcript line (segments[], each ~5s of speech) within ±extend of [t0, t1].
  2. Ask Claude (single batched claude -p call across all spans) to pick a new t0 that starts at a sentence start and a new t1 that lands on a sentence end. New values must be transcript-line boundaries within the window.
  3. Cache on input mtimes; out-of-window or too-short adjustments are rejected and the original [t0, t1] is kept.

Per-span bookend_note records Δt0 / Δt1 and Claude's reason. Pairs well with tighten-pace downstream, which removes dead air inside the re-bookended clip.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.