transcribe
Skill1 starsChanged 21 days agoMITView on GitHub
Turn an audio or video file the user owns into a timestamped transcript, on their own computer with faster-whisper (free, nothing uploaded). Writes a readable transcript with [mm:ss] stamps, plus SRT or VTT captions and a JSON file other skills can read, and walks the user through the install if faster-whisper is missing. Use it for ANY request to transcribe, caption or get the words of a recording, including when the user pastes a YouTube, Instagram or TikTok link: this skill knows how to handle links safely (it never downloads from platforms and shows how to export your own file). Triggers: "transcribe this", "get the transcript of my podcast / webinar / interview / reel", "make subtitles", "SRT file", "what did I say at minute 12", or before talking-head-cards, chapters or clips.
- Installs packages
47 other projects use a version of this.
Adopted by
Most recently taken up by
- linxule/carreltheir own version, first seen Nov 8, 2026
- overment/skillstheir own version, first seen Nov 8, 2026
- TserenTserenov/FMT-exocortex-templatetheir own version, first seen Nov 2, 2026
- iwesys/IWEtheir own version, first seen Nov 1, 2026
- thesammykins/skillstheir own version, first seen Nov 1, 2026
- DojoCodingLabs/remotion-superpowerstheir own version, first seen Nov 1, 2026
Used with
Projects with it also use spreadsheet, think and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
JetBrainsSkillspreadsheet
Use when tasks involve creating, editing, analyzing, or formatting spreadsheets (`.xlsx`, `.csv`, `.tsv`) with formula-aware workflows, cached recalculation, and visual review.
Also in 3 projects that use this
vltingSkillthink
Lightweight plan-to-execution pipeline. Explores codebase, runs council deliberation, presents plan for approval, then executes via /run.
Also in 3 projects that use this
robzolkosSkilldone
Reset the working directory for the next task by ensuring no uncommitted changes exist, then switching to the main branch and pulling latest.
Also in 3 projects that use this
JaimeGBASkillingest
Digest an external source (article, paper, transcript, URL, book chapter) and integrate its content into the knowledge base
Also in 3 projects that use this
AnthropicSkillonboarding
Generate an onboarding checklist and first-week plan for a new hire. Use when someone has a start date coming up, building the pre-start task list (accounts, equipment, buddy), scheduling Day 1 and Week 1, or setting 30/60/90-day goals for a new team member.
Also in 3 projects that use this
MicrosoftSkilldiagnose
Use to systematically diagnose why scenarios fail — identifies root causes by analyzing agent traces, codebase, and behavioral patterns
Also in 3 projects that use this
Other versions47 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 47
- openai/skills★ 27ksame name
- gnekt/My-Brain-Is-Full-Crew★ 3.9ksame name
- badlogic/pi-skills★ 2.5ksame name
- greatSumini/cc-system★ 438same name
- JetBrains/skills★ 386same name
- bholmesdev/skills★ 179same name
- DojoCodingLabs/remotion-superpowers★ 153same name
- thiagofinch/mega-brain★ 95same name
- TserenTserenov/FMT-exocortex-template★ 63same name
- jftuga/transcript-critic★ 40same name
37 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository11 other files it gives its agents.
- find-modelSkill
- generate-imagesSkill
- generate-videoSkill
- humanizeSkill
- launch-videoSkill
- talking-head-cardsSkill
- youtube-auditSkill
- youtube-commentSkill
- youtube-planSkill
- youtube-seoSkill
- youtube-shortsSkill
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 40 edited versions we read keep these.
7
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 33
Ask for, or find, the file path.
- Distinctive: line 46
When you say no to a link or someone else's video, keep it short and complete:
- Distinctive: line 50
Do not point them to another way to get that content (YouTube's transcript panel, a download site, pasting it in), and do not offer to …
- Very specific: line 61
If faster-whisper is missing, the tool prints the install steps.
- Distinctive: line 66
If faster-whisper is installed in a separate Python (a venv), run the tool with that Python.
- Distinctive: line 99
Do not "clean up" what the speaker said into something they did not say.
- Distinctive: line 111
Never invent words for parts that are silent or unclear.
- Distinctive: line 113
Ask before installing anything.
“Common” is measured: kept by at least 6 in 10 of the 40 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
transcribe
Make a transcript with timestamps from a recording the user has as a file. Everything runs
locally, so private calls and unreleased videos never leave the machine. The tool is
scripts/transcribe.py (standard library; it calls faster-whisper when installed).
What this skill reads and writes
- Reads: one audio or video file in the project (often
media/raw/). The founder-context business file only if the user wants names spelled a certain way. - Writes:
media/transcripts/<name>.md(readable),<name>.json(segments, for other skills), and<name>.srt/<name>.vttwhen captions are asked for. - Next:
talking-head-cardsfor graphic cards over the clip, the linkedyt-chaptersoreditskills,youtube-shortsto find clips, or a post-writing skill to repurpose it.
1. Check the source is a file they have
Ask for, or find, the file path. Then make sure of two things.
- It is a file on this computer. If the user gives a YouTube, Instagram, TikTok or other
link, do not download it, not with yt-dlp, a browser, or anything else. Downloading breaks
those platforms' terms, and someone else's video is not theirs to transcribe. Explain that
in one line, then tell them how to get their own copy:
- YouTube: YouTube Studio, Content, the video's menu, Download (their own uploads).
- Instagram: Settings, Your activity, Download your information, or the original file.
- TikTok: their own post, Share, Save video. Or the original from their phone.
- Zoom, Riverside, Descript and phone recorders: the export or the original file.
- It is theirs, or they have permission. If they say it is someone else's video, say you can only work on recordings they made or have rights to.
When you say no to a link or someone else's video, keep it short and complete:
- One line on why: you only work on files the user owns or exported, and downloading from platforms breaks their terms.
- Do not point them to another way to get that content (YouTube's transcript panel, a download site, pasting it in), and do not offer to study or rework someone else's script.
- A way forward with their own material: the export steps above for their own videos, and
an offer to help write an original script with a named skill (the linked
yt-script).
2. Check the setup
python3 <skill>/scripts/transcribe.py doctorIf faster-whisper is missing, the tool prints the install steps. Walk the user through them
and say what to expect: a one-time model download (about 0.5 GB for small, about 3 GB for
large-v3), and that it runs fine on a normal laptop CPU, just slower. Do not run pip
install yourself without asking, because it changes their machine. Offer to do it.
If faster-whisper is installed in a separate Python (a venv), run the tool with that Python.
3. Pick the settings
- Model:
smallis the default and is good for clear speech. Usemediumorlarge-v3for accents, noise, or several speakers, and say it will take longer. Useturbowhen the user wants it fast. - Language: leave it out to detect. Set it (
--language ar,--language es) when you know it, since detection can slip on short clips or mixed languages. - Formats:
md,jsonby default. Addsrtorvttfor captions.--wordskeeps word timings in the JSON for tools that sync to single words (the SRT and VTT options turn it on).
4. Run it
python3 <skill>/scripts/transcribe.py run media/raw/episode-12.mp4 --formats md,json,srt --out media/transcriptsA long recording can take a while on a CPU. Tell the user before you start. If it fails,
read the message: a damaged file or one with no audio track is the usual cause. If it finds
no speech, try the right --language or a bigger model.
To make captions later from an existing JSON, without the model:
transcribe.py convert media/transcripts/episode-12.json --formats srt,vtt
5. Hand it back
Show the file paths and a short preview (the first few [mm:ss] lines from the .md). Then
say plainly: it is a machine transcript, so check names, numbers and quotes before publishing
any of them. Say this even before a run, when you explain what they will get.
Whisper often misspells names and brand words. If the business file lists the right
spellings, offer to fix them in the transcript, and only change what you are sure of.
Do not "clean up" what the speaker said into something they did not say. Fixing a misheard word is fine. Rewriting a sentence is a different job, and it must be marked as an edit.
End with one line naming the next skill that fits, by name: talking-head-cards for text
cards over the clip, youtube-shorts to find clips, youtube-seo or the linked yt-chapters
for chapters, or a post-writing skill.
Rules
- Files only. Never download from any platform or site, even when asked, even "just this once".
- Local only. Do not send the audio to a paid API from this skill. If the user wants a hosted service, say it costs money and uploads their audio, and let them choose.
- Never invent words for parts that are silent or unclear. Leave the gap.
- Say it is a machine transcript every time.
- Ask before installing anything.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.