agentleFS
titouannwtttitouannwtt/claude-tools/skills/video-transcript/SKILL.md

video-transcript

Skill1 starsChanged 12 days agoMITView on GitHub

Get the transcript of a YouTube video from its URL, to read it, summarize it or integrate it into a project. Use when the user gives a YouTube link and asks what is said in it, its transcript, its subtitles, or an analysis of its content.

  • Reads credentials
  • Installs packages

20 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 12 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the script fails without an expected code (1), or if an output is inconsistent:

    line 35
  2. Distinctive:

    Read the full error message.

    line 36
  3. Distinctive:

    Never display the proxy value, neither in the chat nor in a versioned file.

    line 52
  4. Very specific:

    Only with a cookie file the user supplied for that purpose.

    line 56
  5. Distinctive:

    Never display the content of a cookie file.

    line 60
  6. Very specific:

    Read the .txt, then answer the user's question from the text, citing the useful timestamps ([00:12:40]).

    line 64

“Common” is measured: kept by at least 6 in 10 of the 12 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

video-transcript

Objective: turn a YouTube video into timestamped text, then understand it from that text. The script does all the work (zero model tokens). Your role: run it, read the result, repair the script if needed.

Commands

Script: ~/.claude/skills/video-transcript/scripts/transcript.py. It always writes JSON to stdout.

  1. Route A, existing subtitles (try first, fast and no video download): python3 <script> subs "<url>" --lang fr,en --out <folder>
  2. Route B, audio then local transcription (if route A found nothing): python3 <script> audio "<url>" --lang fr --model small --out <folder>
  3. Audio file already present: python3 <script> transcribe <file> --out <folder>
  4. Check the dependencies: python3 <script> doctor

Exit codes and reaction

CodeMeaningAction
0SuccessRead the .txt (lines [hh:mm:ss] text) or the .json (segments).
3No subtitlesRun route B (audio).
4Download blocked (429, 403, "not a bot", blocked IP)1. Update yt-dlp (brew upgrade yt-dlp or pip install -U yt-dlp): a 403 often comes from an outdated version. 2. Rerun. 3. If it still blocks, rerun only once with --use-proxy. Never use the proxy without a block.
5Missing dependencyInstall what the message indicates (pip install -U yt-dlp, pip install faster-whisper, ffmpeg).
2Invalid URL or fileAsk for the correct URL.
1Unexpected errorRead the message, fix the script, rerun (see Repair).

Repair

If the script fails without an expected code (1), or if an output is inconsistent:

  1. Read the full error message.
  2. Fix the script in <claude-tools>/skills/video-transcript/scripts/transcript.py (the symbolic link points to it), and run python3 -m unittest discover -s tests in the skill's folder.
  3. Rerun the command. Two attempts at most, then report the problem to the user with the exact message.

Installing the dependencies

  • yt-dlp and ffmpeg: Homebrew (Mac) or system packages (VPS).
  • faster-whisper: on the Mac, in a dedicated environment ~/.venvs/video-transcript (Homebrew's Python, PEP 668). The script switches to that environment by itself. On the VPS, pip install --user faster-whisper.
  • In both cases, av must stay below version 16 (pip install "av<16"): recent versions remove a parameter that faster-whisper uses.

Proxy (fallback)

  • The IPRoyal proxy is used only with --use-proxy, and only after a block (code 4).
  • Its address comes from local.env at the root of claude-tools (key IPROYAL_RESIDENTIAL_URL, the residential proxy; ignored by git). Do not use the Web Unblocker: it intercepts TLS and breaks yt-dlp. Limited budget (2 GB): prefer subs with the proxy.
  • Never display the proxy value, neither in the chat nor in a versioned file.

Cookies (when YouTube asks to sign in, "not a bot")

  • Only with a cookie file the user supplied for that purpose. Make a private copy (mode 600) limited to the YouTube and Google domains, outside any repository, and pass it with the environment variable YTDLP_COOKIES=<copy> (yt-dlp rewrites the file: never point it at the user's original, one copy per parallel process).
  • With cookies, subtitles still fail (429, or a PO Token is required): go straight to route B (audio), which works.
  • Never display the content of a cookie file; delete the copies when the batch is done.

After the transcript

Read the .txt, then answer the user's question from the text, citing the useful timestamps ([00:12:40]). If the video is long, summarize in sections rather than copying everything.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.