Transcript Formatter

The Transcript Formatter is a Claude skill that turns an unreadable wall of transcript text into a clean markdown document with section headers and paragraph breaks. Every single word stays exactly where it was. Nothing rewritten, nothing dropped.

The Problem

You pulled the transcript. Maybe it's a YouTube auto-caption dump, maybe a Whisper export, maybe an hour-long Zoom call that Otter spat back at you. Either way, what you're staring at is 11,000 words with no punctuation, no paragraphs, no breaks. One unbroken block from "so today I want to talk about" all the way to "thanks for watching."

You try to read it. You lose your place by line three. You scroll, hunting for the part where they actually said the thing, and there's no structure to grab onto, so you scroll past it twice. You think about pasting it into ChatGPT to clean up. Then you remember what that does, it "tidies" the text, quietly swaps words, smooths out the "um"s, and now your transcript isn't a record anymore. It's a paraphrase you can't trust.

So it just sits there. Too long to read, too valuable to delete, too risky to let an AI rewrite.

This skill ends that standoff. It adds the headers and paragraph breaks that make it readable and proves it changed not one word.

What this Claude skill does

• Breaks the wall into paragraphs: blank-line breaks at natural pause points, 2 to 6 sentences each, the way a person would actually speak.
• Preserves every word exactly: no word added, removed, reordered, or altered. No punctuation invented, no "um"s scrubbed, no contractions expanded.
• Adds ## section headers at topic shifts: chapters, life stages, projects, agenda items, Q&A turns, so you can skim the structure in seconds.
• Verifies word-for-word before writing: a deterministic checker compares the formatted output against the original word sequence and aborts if anything drifted.
• Handles any length: a chunk-then-parallel pipeline splits long recordings into pieces, formats them at once, and stitches them back, so a two-hour podcast doesn't blow the context window.
• Works on any source: YouTube captions, podcasts, Whisper/Descript exports, Zoom/Meet/Teams/Otter meeting recordings, interview transcripts.

Why it's amazing!

• Read an hour-long transcript in the time it takes to skim, not slog.
• Trust the output as a real record, because the verifier guarantees no word changed.
• Throw a two-hour recording at it without hitting a length wall.

Who it's for

Anyone sitting on a transcript they can't stand to read. Podcasters cleaning up episode transcripts, founders turning a YouTube talk into something usable, researchers working through interview recordings, teams that need meeting notes from a raw Otter export, and content creators prepping a long video for repurposing. If you've ever pulled a transcript and immediately wished it had paragraphs, this is for you.

How it works

Hand Claude a transcript file and ask:

• "Format this transcript so it's readable."
• "Add headers and paragraphs to @00-inbox/transcript-my-podcast-ep-12.md"
• "Clean up this YouTube transcript, don't change any of the words."
• "Section this meeting transcript into the agenda items they covered."
• "I've got a 90-minute interview transcript that's one giant block, make it skimmable."

You hand it a block that opens like this:

so today i want to talk about how i failed 27 startups before anything worked and i'll be honest the first one was in college we called it tinder for sports nobody used it and then there was the rock bottom moment where i had like 12 dollars in my account

And get back structured markdown:

First Startup: Tinder for Sports

so today i want to talk about how i failed 27 startups before anything worked and i'll be honest the first one was in college we called it tinder for sports nobody used it

Rock Bottom

and then there was the rock bottom moment where i had like 12 dollars in my account

Same words, same order, same casual lowercase. The only things added are the headers, the paragraph breaks, and the occasional horizontal rule between sections. Before the file is written, a word-level verifier confirms the formatted text matches the original word-for-word. If any chunk drifted, it names the chunk and refuses to overwrite your file.

Pairs well with these other skills

Formatting the transcript is the cleanup step. These skills get you the raw text in the first place, or turn the readable version into something you publish:

• Content Repurpose Generator: feed it the cleaned-up transcript and get 20 to 30 platform-native assets back, posts, threads, a newsletter, short scripts.
• Video Scripts and Captions Generator: turn the structured transcript into new scripts and captions for Shorts, Reels, and long-form.
• Voice Cloner: point it at a readable transcript to extract how the speaker actually talks, then write new content in that voice.

FAQ

Does it cost anything to run?

Only a Claude subscription, no other external APIs needed to run this skill.

Will it change any of my words?

No. It only inserts blank lines, ## headers, and horizontal rules. A word-level verifier checks the output against the original before writing and aborts if even one word drifted.

Does it summarize or shorten the transcript?

No. It formats the full text. If you want a shorter or paraphrased version, that's a different job, this skill keeps every word.

How long a transcript can it handle?

As long as you've got. It splits the transcript into roughly 1,500-word chunks, formats them in parallel, and stitches them back, so a two-hour recording doesn't hit the context limit.

Does it work on transcripts that aren't from YouTube?

Yes. Podcasts, Whisper and Descript exports, Zoom/Meet/Teams/Otter meeting recordings, interview transcripts, any wall-of-text transcript regardless of source.

Will it fix punctuation and capitalization?

No. It deliberately leaves the words exactly as written, including lowercase auto-caption text and filler words. Adding punctuation would mean altering the text, which it won't do.

What happens if the formatting goes wrong?

The verifier catches it. If a chunk drifted, the skill names the offending chunk and the first divergent word and leaves your original file untouched so you can re-run that piece.

Does it remember past transcripts?

No. Each run formats the file you hand it. There's no stored history across runs.

Safe To Install

Please read before installing. Any skill you copy online is someone else's code running inside your Claude, and a lot of what's out there should give you pause. Most authors who give skills away don't give a second thought to quality or security. Most are vibe-coded slop pushed out untested, and some are the occasional bad actor hiding an exploit in plain sight. Every skill we publish is vetted by a programmer with 20+ years of professional experience and checked line by line for anything that touches your files without reason. With Stim-Pack Studios Claude Skills, you get peace of mind knowing you're installing safe, vetted code, not a stranger's guess.

Add it to your Claude workflow

If you keep pulling transcripts and never reading them, this one earns its slot. It collapses the gap between a raw recording and a document you can actually skim, without putting a single word at risk.

Get the Transcript Formatter Claude skill and turn your next wall of text into something you'll actually read.