🎬 Paste One Link, Get the Whole Transcript — Playlists Welcome

Stop scrubbing back through hour-long videos to take notes. Paste a public YouTube link and you get a readable full transcript — with the messy rolling duplicates of auto-captions scrubbed out — plus a chapter-by-chapter summary in your language, saved as files on your own machine.

• One playlist URL batch-extracts an entire series — blocked videos are skipped, with reasons reported in a table
• Chapter-aware structured summaries + quote search across multiple videos in Video title [MM:SS] format
• 4 output formats — txt, timestamped md, SRT, VTT — with unlimited use inside your own Claude Code

📦 Real content preview

This is a new listing with no reviews yet. So instead of sales copy, here is the actual content of the files you will receive (excerpts translated from the Korean source; commands and code shown as-is). Judge the quality with your own eyes.

① SKILL.md operating rules — "batch processing" is not just a bullet point (verbatim, translated)

• Never paraphrase or edit the transcript source text. Quotes stay in the original language; only summaries follow the conversation language.
• When some videos in a playlist batch fail (no subtitles, restricted access), skip them and report the reasons in a table at the end.

Two blocked videos in a 12-video playlist will not stop the job. When it finishes, you get a table of what was skipped and why — and a rule explicitly forbids the transcript from being quietly "polished." If you quote from videos, these two lines are the whole point.

② SKILL.md summary step — summary quality pinned down by instruction (verbatim, translated)

5. Summarize: If <id>.json contains chapter data, split the transcript at chapter boundaries and summarize in a "Chapter title — summary — start timestamp" structure. If there are no chapters, create topic-level subheadings and summarize. Summaries follow the conversation language; transcript quotes stay verbatim.

Instead of hoping the AI "summarizes it nicely," the behavior is defined separately for chaptered and chapterless videos — so you get the same summary structure every time.

③ Troubleshooting guide — failure modes designed for in advance (2 of 9 symptoms, verbatim, translated)

Symptom	Cause	Fix
All extractions suddenly fail	YouTube-side change broke an outdated yt-dlp	Run yt-dlp -U or pip install -U yt-dlp, then retry
json3 download fails, only vtt available	Some caption tracks do not offer json3	fetch_subs.py falls back to vtt automatically — clean_transcript.py parses vtt too

If YouTube will not serve a caption format, the script falls back to another one on its own; and the single most common breakage — a YouTube-side change — has its first response already decided. The full table has 9 rows like these.

📋 What's included (actual counts)

The scripts use only the Python standard library, and they never download video or audio files — captions and metadata only.

💬 3 usage scenes

1) A lecture series organized in one afternoon

Me: "Extract transcripts for this whole playlist and summarize each video chapter by chapter."
Claude: Extracts 12 videos in sequence → skips 2 with no captions (reasons reported in a table) → generates 10 md files, each structured as "Chapter title — summary — start timestamp"

2) Research that cuts across multiple videos

Me: "Compare what these 3 videos say about 'the limits of RAG.'"
Claude: A table of shared claims and differences, each row backed by a Video title [12:40]-style quote — quotes kept in the original language, never translated or paraphrased

3) Subtitle file production + follow-up questions

Me: "Clean up this video's English captions into an SRT. Also, where does the speaker talk about pricing?"
Claude: Saves a de-duplicated .srt file + quotes the relevant passage with timestamps from the transcript kept on disk

Because transcripts live as files on your machine, you can come back days later and ask, "Find where they mentioned X in that transcript from last time."

🙆 A great fit if you…

• Want lectures, conference talks, and podcasts turned into searchable text you keep
• Are a researcher, writer, or journalist who needs to compare claims across videos with timestamped quotes
• Repurpose video content into blogs and newsletters and need accurate, citable sources
• Are tired of per-conversion fees and credit meters on converter websites

🙅 Not for you if you…

• Want text from videos with no captions at all — this skill does not do speech-to-text (STT)
• Need private, login-required, or age/region-restricted videos — YouTube's access limits make this impossible
• Do not use Claude Code — this product is a skill that runs on top of Claude Code
• Expect mis-transcribed words in auto-captions to be corrected — it never invents content that is not in the source captions

❓ Trust FAQ

Q1. How is it delivered, and how do I use it?
After purchase, download the skill files (SKILL.md + 3 scripts + 1 guide) and drop them into your Claude Code skills directory — that's it. Then paste a YouTube link into a Claude Code conversation and say "get the transcript." No separate server, no sign-up.

Q2. What about updates?
Your files are local, so they keep working. When extraction breaks due to a YouTube-side change, the fix is almost always updating the free open-source engine yt-dlp (yt-dlp -U) — and the skill is designed to suggest exactly that first. Improved versions of the skill itself follow the platform's re-download policy.

Q3. Can I get a refund?
Refunds follow the marketplace's (Capafy's) refund policy. Before buying, please check the "Not for you" section and the supported scope above.

Q4. Do I need an AI subscription? And what languages does it handle?
The skill runs inside Claude Code, so you need a working Claude Code setup. The only extra dependency is the free open-source yt-dlp (brew install yt-dlp or pip install yt-dlp); the Python scripts use only the standard library. On languages: this is a language-agnostic tool. It extracts captions in whatever language the video provides (you can request specific tracks, e.g. --lang en ko); transcripts and quotes always stay in the video's original language, while summaries and answers are written in your conversation language — chat in English, get English summaries. The skill's internal instructions are written in Korean, which Claude reads natively; it does not affect your experience in English.

Q5. Can I use extracted transcripts commercially?
The skill (the tool itself) is yours to use freely in work and commercial projects. However, copyright in the extracted transcripts belongs to the original video creators — complying with their rights and YouTube's terms when you quote or repurpose the output is your responsibility.

💳 Purchase info

$3 · one-time payment · unlimited use

No per-use fees, no monthly subscription, no credit meter. Buy once and run it in your own Claude Code on one video or a 100-video playlist, as many times as you like. For the price of a coffee, stop re-watching videos to take notes — starting today.