One long video becomes clips, chapters, and an article

A long video is the most expensive thing most teams make, and it usually ships once. This loop reads the transcript and turns it into the pieces people actually find: chapters for the video page, short clips for the feeds, an article for search, and posts that point back.

Use it when you publish anything long: a YouTube episode, a webinar, a talk, a product walkthrough.

You need a transcript with timestamps. Transcription happens outside this loop, and a transcript without timestamps will fail the verifier on purpose.

You get a chapter list, a ranked list of clip picks with start and end times, an article draft, and social post drafts, each quote carrying the timestamp it came from.

The gate is you. It drafts into a folder or a pull request. It never uploads, posts, or edits the video.

What makes a clip worth cutting

The clip picks are where this loop earns its keep. A moment qualifies when it:

• Makes sense cold. Someone who has not seen the rest of the video understands it.
• Has one point. Not a setup and a payoff separated by a tangent.
• Starts on a strong line. The first sentence is the hook, not a filler word.
• Ends clean. It finishes the thought instead of trailing into the next topic.

A moment that needs context from earlier in the video is an article paragraph, not a clip.

Goal prompt

Read <transcript-file>. It has timestamps. Read <voice-file> if it exists.

1. Write chapters: one line per topic shift, each with the timestamp where
   the topic starts. The first chapter starts at 00:00.
2. Pick clip candidates: moments that make sense with no prior context, carry
   one point, start on a strong line, and end on a finished thought. For each,
   record start time, end time, the opening line verbatim, and one sentence on
   why it stands alone. Rank them.
3. Draft one article built from the transcript. Every quotation in it is
   copied verbatim from the transcript and followed by its timestamp.
4. Draft one social post per top-ranked clip, each quoting the speaker
   verbatim with the timestamp and linking to <video-url> at that time.

Write everything to <output-folder>. Never paraphrase inside quotation marks.
Never add a claim, number, or example the speaker did not say. If a section of
the transcript is unclear, mark it unclear rather than guessing the words.

Verifier prompt

Open the transcript and every file in the output folder, in a fresh session.

FAIL if any text inside quotation marks does not appear verbatim in the
transcript at the timestamp given.
FAIL if any quotation has no timestamp.
FAIL if any clip has an end time before its start time, or falls outside the
video's length.
FAIL if any chapter timestamp is out of order.
FAIL if any number or named claim in the article or posts is absent from the
transcript.
FAIL if anything was uploaded, posted, or scheduled.

Otherwise PASS. On FAIL, print the file, the quote or line, and the reason.

Stopping condition

Chapters, ranked clip picks, one article draft, and one post per top clip exist in the output folder, every quotation matches the transcript verbatim at its timestamp, and the verifier printed PASS.

What it may and may not do

May: read the transcript and voice file, write drafts into the output folder, open a pull request.

May not: paraphrase inside quotation marks, invent a claim the speaker did not make, upload or cut video files, post to any account, or edit the published video.

What this will not tell you

It picks clips from the words, not the picture. A line that reads well can land badly on camera, and a great visual moment with no strong line will be missed. Watch the top picks before you cut them.

It also cannot tell you whether the video was worth repurposing. If nobody watched the long version, more pieces of it will not fix the topic.