Claude Code YouTube Skill: From Transcripts to Research Notes

A Claude Code YouTube skill packages instructions for working with YouTube content: fetching a transcript through the YouTube API or a transcript tool, summarizing it, extracting key points, or structuring research notes from it. You activate the skill, supply a video URL or ID, and get structured output rather than a raw wall of text.

The quality of this output depends almost entirely on the accuracy of the transcript. Auto-generated captions have errors, especially on technical or domain-specific content. fenbs can hold the video links, your reviewed notes, and follow-up tasks, so the work you do on a transcript does not disappear when the session ends.

How a YouTube transcript skill works

YouTube exposes transcript data through its API and, for many videos, provides auto-generated captions. A skill that works with transcripts either calls the YouTube API directly (which requires an API key configured in your environment) or uses a library such as youtube-transcript-api in Python to fetch available caption tracks.

Not all videos have accessible transcripts. Transcripts may be disabled by the creator, unavailable in the language you need, or exist only as auto-generated captions that are inaccurate for specialized content. Check the skill's documentation for how it handles a video with no transcript rather than discovering this mid-workflow.

Some skills fetch a full transcript; others chunk it to fit the context window. If the skill chunks the transcript, check how it handles content that spans chunk boundaries. Key points should not be cut off because they happen to fall at a chunk edge.

Evaluating a YouTube skill before you rely on it

Read the skill instructions before running it on important content. Look for answers to these questions: Does it call any external service beyond YouTube? Does it cache or store the transcript text anywhere? What does it do when the transcript is unavailable? What format does it produce?

Test the skill on a video you know well. Compare the skill's output to your own notes on that video. Look for names, figures, or key claims that appear in the video but not in the skill's output — these indicate that the transcript missed or misheard them. Check a technical or proper noun in the transcript against what the speaker actually said.

• Confirm the transcript language matches the video language before using the output.
• Spot-check five specific claims from the output against the video at their timestamps.
• Note any section the skill marked uncertain or skipped and verify those manually.
• Check whether proper nouns, acronyms and technical terms are rendered correctly in the transcript.

Using transcripts responsibly

A YouTube transcript is the creator's words. Using it for research, summarizing for personal notes, or extracting factual points for a project is generally consistent with fair use. Republishing a substantial portion of the transcript, building a product on top of it without permission, or presenting auto-generated captions as verified quotes are different matters.

Auto-generated captions are not the creator's verified text; they are a machine transcription that may contain errors. When you quote or cite a video, quote the video itself rather than the auto-generated transcript. Verify important claims directly against the video before including them in a report or publication.

YouTube's terms of service govern automated access to its platform. If your skill uses the YouTube Data API, stay within the quota and usage policies documented at YouTube's developer site. Using unofficial scraping tools rather than the official API may violate those terms.

Turning transcript notes into useful research

A raw transcript summary is a starting point, not a finished research note. Before using the output, mark which claims you have verified against the video, which are paraphrases from a machine transcript, and which require a primary source check. If you are building an argument on a claim from a video, watch that section of the video yourself.

Structure the output as a set of actionable items: what was confirmed, what needs a primary source, and what questions came up during review. This makes the note useful to a future session or a colleague, rather than a dump of auto-generated text.

How to do this with fenbs

1. Create a free account, free for one person with no credit card, and add a project for the research.
2. After the skill produces output, review it against the video and note what was verified and what was not.
3. Add the video URL to Links and listings in fenbs with a short summary of what was useful and what needs checking.
4. Create a task for each claim or section that requires a primary source or further verification.
5. Connect Claude Code to fenbs through Integrations and ask it to read the project notes before continuing research in a new session.

fenbs does not fetch YouTube transcripts, call the YouTube API, or store video content. The skill handles the transcript work; fenbs holds the links, the checked notes, and the open tasks. If you are only extracting notes from one video and will not return to the research, a local note may be enough without a fenbs project.

For related reading, see Claude Code sandbox for notes on running external API calls in a controlled environment, and Claude skills examples for other skill patterns.