youtube
/youtube
Czyta film z YouTube: tytuł, kanał, rozdziały i transkrypcję ze znacznikami czasu, a potem odpowiada na pytania o treść.
- Autor
- teenxgrailed ↗
- Repozytorium
- needt-plugin ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 06.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 2
- Skrypty
- 1
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 1
Brak sygnałów. Repozytorium zmienione 06.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 1 skrypt w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/transcript.py
Co robi i kiedy
Czyta film z YouTube: tytuł, kanał, rozdziały i transkrypcję ze znacznikami czasu, a potem odpowiada na pytania o treść.
Wyzwalacze z opisu
Oryginalny opis (EN)
Reads a YouTube video (title, channel, date, chapters, full timestamped transcript), then answers from it. Fires on any youtube.com, youtu.be or Shorts link, whatever else the message asks; the link is read first. Use for "summarize this video", "what does this video say", "transcript", "/needt:youtube", "перескажи видео", "что в этом видео", "транскрипт", "про що відео", "транскрипція", "was steht im Video", "Transkript", "resume el vídeo", "transcripción", "résume la vidéo", "transcription". Not for finding videos (search).
Jak działa
- Setup check
- A link anywhere means: read it first
- Used by other skills — the contract
- 1. Get the transcript — first route that works
- 2. Read it
- 3. Answer
Podgląd SKILL.md
---
name: youtube
description: Reads a YouTube video (title, channel, date, chapters, full timestamped transcript), then answers from it. Fires on any youtube.com, youtu.be or Shorts link, whatever else the message asks; the link is read first. Use for "summarize this video", "what does this video say", "transcript", "/needt:youtube", "перескажи видео", "что в этом видео", "транскрипт", "про що відео", "транскрипція", "was steht im Video", "Transkript", "resume el vídeo", "transcripción", "résume la vidéo", "transcription". Not for finding videos (search).
---
# Needt · YouTube
The job: understand what the video says, as if you had watched it, and answer the user's real
question about it, with timestamps. A summary of the title and description is not that. A
transcript dumped into the chat is not that either.
## Setup check
No line starting with `Needt ·` in this conversation yet → run [setup](../setup/SKILL.md),
print its line first and continue in the same reply; the `YouTube ·` header below is the
first line of this skill's own report, after it. Present → reuse it.
## A link anywhere means: read it first
A YouTube link in a message is input, not a request. Whatever the message asks — «сделай
задачи из этого видео», "is the plan in this talk any good?", "compare this with Notion",
"what do people say about this?" — the video is read with section 1 before anything else, and
the skill that owns the request (needt, critic, partner, search, prompt, asking) works from
the transcript file. The user never has to say "watch it" or name this skill. Without a link,
the skill fires only on the phrases in its description.
## Used by other skills — the contract
Who calls this skill and what comes back: [skills map](../setup/references/skills-map.md).
Any Needt skill that meets a video calls this one in place (no hand-off, no user prompt):
1. Run section 1 (first route that works) → a file `<out>/<id>.md` and the header line
`YouTube · <title> · <channel> · <duration> · captions: <lang> <manual|auto>|none`.
2. Print the header line once in the reply, then read the file and continue with the
caller's own rules. Do not produce the section-3 summary unless that is what the user
asked for.
3. Several links → one run per link, in parallel; the files stay for the whole conversation,
so a later question about the same video is answered from the file, never refetched.
4. Nothing worked → the caller says in one line which routes failed and goes on with what it
has (title, description) labelled as unread.
Who calls it for what:
| Caller | The video is… | What it takes from the transcript |
|---|---|---|
| [needt](../needt/SKILL.md) | a how-to, a checklist, a plan to follow | the steps in order → tasks, each with the `[mm:ss]` in the description |
| [search](../search/SKILL.md) | one of the sources it picked | tools, numbers, claims with timestamps, cross-checked against other sources |
| [critic](../critic/SKILL.md) | a plan, a strategy, a pitch | the plan as stated, then the normal critique |
| [partner](../partner/SKILL.md) | a competitor, a method, an idea the user brings | what is actually claimed vs what the user's data says |
| [prompt](../prompt/SKILL.md) | material for the prompt | the key points into `<inputs>`, cited with timestamps |
| [asking](../asking/SKILL.md) | a fact a question depends on | the fact, looked up, never asked |
| [setup](../setup/SKILL.md) | — | lists which transcript routes exist here |
## 1. Get the transcript — first route that works
1. **Bundled script (shell available).** It sits next to this file:
`scripts/transcript.py` (use the base directory the host gave for this skill).
```bash
python3 -I <skill-dir>/scripts/transcript.py "<url or id>" [--lang ru,en] [--out DIR]
```
It writes `<out>/<id>.md` (default: the system temp dir, `needt-youtube/`) with the metadata,
chapters, description and the transcript in ~30-second paragraphs `[mm:ss] …`. Always run it with `--print head` and `--out` pointing at a scratch folder: stdout then
holds only the header, chapters and a `FILE <path>` line, and you read the file in
pieces (a long transcript dumped into the chat is the failure this guards against).
`--print all` only for videos under ~3 minutes. Without `--lang` it takes manual captions in the video's own language, then the
original auto track. Pass `--lang` only when the user wants a specific language.
Exit codes: `2` — no `yt-dlp`: offer `brew install yt-dlp` (macOS) or `pipx install yt-dlp`,
run it on a yes, retry; `3` — private, removed or age-gated video: say so; `4` — the video
has no captions in any language: say so and work from the description and chapters only,
labelled as such.
2. **Firecrawl** (`firecrawl_scrape` with `alexandria`, provider `youtube-com`):
`videos/video_details` and `videos/video_transcript` with `{video_id}` in one call
(the field takes an array). ~5 credits each.
3. **A youtube-transcript MCP** (`youtube_get_transcript`). A 401 or "IP blocked" means
YouTube blocks that scraper from this network — move on, do not retry.
4. **Nothing worked** → say which routes were tried and why each failed. Never describe a
video from its title, thumbnail or memory as if it had been read.
Several videos: run the script once per video in parallel; more than three long ones go to
Sonnet sub-agents per [delegation](../setup/references/delegation.md), each given section 2
of this file and the file path.
## 2. Read it
Read the whole transcript, not the first screen. Note for yourself:
- what the video is (tutorial, review, talk, vlog, ad) and who is speaking;
- the main claims and the steps in order, each with its timestamp;
- every tool, product, site and number named, with its context;
- what the author says not to do, and any sponsor segment (skip it in the summary, name it);
- what is shown but not said — a transcript is speech only, so "as you can see here" means
the answer is on screen; say that you could not see it.
Auto captions mishear names, brands and numbers. A name that looks wrong is marked `(?)`
and checked against the description or a quick search before anything depends on it.
## 3. Answer
First line, fixed shape:
`YouTube · <title> · <channel> · <duration> · captions: <lang> <manual|auto>|none`
Then, in the user's language:
- **The user asked something specific** ("what tool does he use", "is X covered") → answer
that directly, with the timestamp(s), then stop or add one line of context.
- **Just a link or "о чем видео"** → one sentence on what it is and for whom; 3–7 key points
with `[mm:ss]`; tools and numbers named; one line on whether it is worth watching in full and
which part, tied to the user's project when one is known.
- **"Транскрипт" / "дай текст"** → give the file path; quote only the passages they need.
Cite as `[mm:ss]` linked to `https://youtu.be/<id>?t=<seconds>` where links render. Keep the
file: follow-up questions about the same video are answered from it without fetching again.
Actionable steps the user wants to do → hand them to [needt](../needt/SKILL.md).
Pliki w paczce (2)
- SKILL.md 7 KB
- scripts/transcript.py 7 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · teenxgrailed/needt-plugin | 100 | MIT |
| AceDataCloud/Skills | 100 | Apache-2.0 |
| PhiDuyAnh/learning-with-ai-kit | 100 | MIT |
| cowork-os/cowork-os | 100 | MIT |
| hackers-reality/youtube-skillbox | 100 | MIT |
| hyperfx-ai/marketing-skills | 100 | MIT |
| baofeng-tech/agent-skills | 100 | MIT |
| windsor-ai/claude-windsor-ai-plugin | 100 | MIT |
| tikhub/tikhub-plugin | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki ac353a28eed8 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100youtube-transcript
Pobiera transkrypcję filmu YouTube jako zwykły tekst ze znacznikami czasu, skryptem Pythona bez zewnętrznych zależności, także z chmury. - 90watch-video
Ogląda film z pliku lub adresu, czytając transkrypcję ze znacznikami czasu i klatki kluczowe, i odpowiada na pytania o treść. - 100youtube-transcript
Pobiera transkrypcję filmu YouTube, gdy użytkownik prosi o tekst, napisy lub streszczenie treści z linku. - 100yt-lokus
Transkrybuje film z YouTube i zapisuje go jako notatkę Markdown z metadanymi, podsumowaniem, oceną i znacznikami czasu. - 100youtube-download
Pobiera transkrypcję, metadane, rozdziały i slajdy z filmu YouTube przez sesję przeglądarki gstack, do dalszej pracy. - 100youtube-summarize
Streszcza film z YouTube w punktach, cytatach i linkach z czasem, odpowiada na pytania o treść i tłumaczy wybrane fragmenty.