Never Rewatch a 30-Minute Video Again — Shorts Candidates Arrive as Production-Ready Cards

• Every candidate is one card: timecode + first-3-seconds hook + vertical-format script + 2 title options with hashtags + ffmpeg cut command. Most of the deliberation is finished in text before you ever open an editing tool.
• Picks are made by criteria, not gut feel. A 9-type hook catalog and a 5-point selection checklist are included as documents, so you can verify the "why this segment" reasoning yourself.
• Transcript extraction and cut-command generation are handled by 2 bundled Python scripts. Paste the output into your terminal and the candidate segments become cut files.

📦 Real content preview

This product has no reviews yet, so here is the judgment material up front. The excerpts below are taken directly from the files you receive (source documents are written in Korean; excerpts translated here).

1) Hook type catalog — from references/hook-patterns.md (4 of 9 types)

Type	Pattern	Example
Contrarian	Open with a claim that flips conventional wisdom	"Working hard is exactly what's blocking your promotion"
Don't-do / warning	Lead with what not to do	"Say this in an interview and you're out"
Result-first	Show the outcome first, make the process the mystery	"This one habit doubled my monthly revenue"
Quote / statement (strong for interviews)	Put the guest's most striking line first	"That was the day I handed in my resignation"

Hook strength rating: exact type match + specificity (numbers, proper nouns) = strong. Type match but flat phrasing = medium. No type, has to open with context = weak (drop the candidate or rearrange).

Beyond these 4, the document covers question, number, relatable-situation, insider-secret, and challenge-proof types (9 total), plus caption emphasis rules ("max 12–16 characters per line, never more than 2 lines on a vertical screen") and a per-platform table (Shorts / Reels / TikTok) of recommended lengths and hook characteristics.

2) The clip card every candidate gets — from SKILL.md, step 3

## Clip N — [working title]
- ⏱ Timecode: 00:12:34.5 → 00:13:21.0 (46.5s)
- 🎯 Selection rationale: (e.g. question-type opening + twist at 15s, stands alone without context)
- 🪝 First-3-seconds hook line: "..." (hook type: contrarian)
- 📝 Vertical-format script: rewritten as opening hook → core → closing CTA
- ✂️ Caption emphasis: words to enlarge (bold), line-break positions, pinned top-of-screen label
- 🏷 2 title options + 5–6 hashtags

3) ffmpeg cut command generation — from SKILL.md, step 4

python3 {skill_dir}/scripts/make_cut_commands.py clips.json --input "source.mp4"            # lossless cut (-c copy)
python3 {skill_dir}/scripts/make_cut_commands.py clips.json --input "source.mp4" --mode precise  # precise cut (re-encode)
python3 {skill_dir}/scripts/make_cut_commands.py clips.json --input "source.mp4" --padding 2     # padded cut
python3 {skill_dir}/scripts/make_cut_commands.py clips.json --input "source.mp4" --script > clips.sh  # bundle into a shell script

The default is a lossless cut (fast); when frame accuracy matters there's a re-encoding precise mode; and when timecodes are shaky because the subtitles were auto-generated, a padded cut adds breathing room on both ends. Three modes for three situations — and when working from auto-generated subtitles, the skill is designed to present the padded version alongside the default.

What's included (actual counts)

How it's used — 3 scenes

Scene 1. Interview channel owner

"Here's the URL for this 52-minute interview. Pull 5 Shorts candidates. Skip anything about the guest's personal life — focus on career advice."

→ 5 candidates arrive as cards (timecode, selection rationale, hook line, emphasis words, 2 title options, hashtags), the personal-life segments that were excluded are listed separately with reasons, and ffmpeg cut commands are attached for all 5 segments.

Scene 2. Course creator (local file)

"I have lecture.srt. Give me 3 segments under 45 seconds that make sense without any surrounding context. Question-style titles."

→ No YouTube needed — the same analysis runs on SRT/VTT files alone. You get 3 candidates that satisfy the conditions (under 45s, standalone context, question-style titles) plus cut commands referencing your local file.

Scene 3. Agency producer

"Client webinar URL. Organize 8 candidates into a rationale table for a report document, and bundle the cut commands into one clips.sh."

→ You get a summary table with timecode, length, selection rationale, and recommended-platform columns, a report-ready Markdown document, and a single shell script collecting every cut command. Because the rationale is explicit, it drops straight into a client proposal.

Buy this if

• You publish long-form consistently, but short-form repurposing keeps slipping because "there's no time"
• You work at an agency or as an editing contractor and need clip selections documented with rationale
• You're already fluent in your editing tool, but resent rewatching footage just to decide where to cut
• Copy-pasting a one-line command into a terminal doesn't scare you

Skip this if

• You expect a finished short-form video file (burned-in captions, 9:16 crop, music, rendering) — this skill delivers judgment, scripts, and cut commands, and stops there
• You mostly work with videos that have no subtitles at all — speech-to-text (STT) is not included; the skill only points you to a local Whisper setup
• You need to process membership, private, or age-restricted videos — anything yt-dlp can't reach is out of scope
• You're trying to buy views or virality itself — selection is a transcript-based judgment; performance depends on editing quality, upload timing, and other external factors

FAQ

Q1. How is it delivered, and how do I use it?
Immediately after payment you download the skill package (SKILL.md + reference document + 2 scripts). Drop it into Claude Code's skills directory and it's auto-recognized — asking "pull Shorts candidates from this video" in chat runs the full workflow.

Q2. What about updates?
Improved versions reflected in the listing can be re-downloaded from the same product page at no extra cost.

Q3. Can I get a refund?
Refunds follow the sales platform's (Capafy) refund policy. Before buying, please check the "Skip this if" list and the content preview above to confirm the scope.

Q4. Do I need AI or other software?
You need Claude Code (the skill's runtime). Subtitle extraction uses yt-dlp (required) and executing cuts uses ffmpeg (recommended) — both free and open source. If either is missing, the skill walks you through installation first. Downloading YouTube subtitles requires a network connection.

Q5. What about language?
The tooling is language-agnostic: the transcript script pulls subtitles in whatever languages the video offers, and ffmpeg commands don't care about language. The guide documents (hook catalog, checklists) are written in Korean — but the skill instructs the AI to serve and adapt all content in your conversation language, so through Claude you'll see everything in English (or any language you use). One honest caveat from the skill itself: hook and copy quality is tuned for Korean and English; other languages may vary.

Q6. Can I use it commercially?
No restrictions on commercial use — your own channel, client work at an agency, and so on. Copyright and platform-terms compliance when reworking other people's videos remains your responsibility, and the skill itself appends that notice to its output.

Purchase info

$3 · one-time payment · unlimited use. Not a subscription. Buy once and use it on as many videos as you like at no extra cost. Compare $3 against the time you normally spend picking segments from a single video — and if the document quality in the preview above didn't meet your bar, don't buy. Showing you the judgment material before purchase is what this page is for.