メニュー
Analyze YouTube videos, web pages (blog posts, articles), or local files (TXT, HTML, PDF) using Google Gemini API. Extracts key takeaways, summaries, code, prompts, and more.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | yt-transcript-summary |
| description | Analyze YouTube videos, web pages (blog posts, articles), or local files (TXT, HTML, PDF) using Google Gemini API. Extracts key takeaways, summaries, code, prompts, and more. |
| triggers | ["summarize youtube video","summarize youtube","summarize transcript","analyze youtube","analyze video","yt summary","yt transcript","youtube takeaways","extract from video","summarize article","summarize blog post","analyze web page","analyze url"] |
Analyze YouTube videos, web pages (blog posts, articles), or local files (TXT, HTML, PDF) using Google Gemini API via the yt_analyze.py CLI tool.
The script and supporting files are co-located with this skill:
~/.claude/skills/yt-transcript-summary/
├── SKILL.md
└── scripts/
├── yt_analyze.py
├── my_prompt.txt
└── requirements.txt
Script dir: ~/.claude/skills/yt-transcript-summary/scripts/
Parse the user's request for these inputs:
| Input | How to pass it |
|---|---|
| YouTube URL | Positional argument: python yt_analyze.py "URL" |
| Web page URL (blog, article) | Positional argument: python yt_analyze.py "URL" (auto-detected as non-YouTube) |
| Local file path (txt, html, pdf) | --file "path/to/file.ext" |
| Custom prompt (inline text) | -p "prompt text" |
| Built-in preset | --builtin <preset> |
| Transcript mode | --transcript (optionally --speakers "Name1" "Name2") |
| Segment analysis (YouTube only) | --start MM:SS --end MM:SS |
| Custom output path | -o "output.md" |
URL and --file are mutually exclusive — the script enforces this.
The script automatically detects the source type:
file_data--file) → reads the file based on extension:
.txt, .md, .json, .csv, etc. → read as plain text.html, .htm → parse HTML tags, extract text content.pdf → extract text from all pages using pdfplumberThe user may provide multiple URLs in a single request, separated by commas, spaces, or newlines. They can be a mix of YouTube and web page URLs. For example:
/yt-transcript-summary https://youtube.com/watch?v=abc, https://example.com/blog-post, https://youtube.com/watch?v=def
When multiple URLs are detected:
python yt_analyze.py "URL" <shared-flags> command for each URL. All other flags (e.g., -p, --builtin, --transcript) apply to every URL.Do not run multiple URLs in parallel — the Gemini API may rate-limit concurrent requests.
--transcript--builtin <preset>-p "their prompt"my_prompt.txt)summary, summary-full, summarize-tests, takeaways, extract-code, extract-python, extract-notebook-code, extract-notebook-cells, extract-prompts, extract-system-prompts, extract-markdown, extract-markdown-contents, extract-source-and-markdown, extract-text, extract-commands, extract-terminal, extract-slides, extract-prompts-and-responses, extract-all, extract-task-instructions, extract-skills, extract-agentic-prompts, extract-claude-md, extract-prd, extract-workflow-md, extract-settings-json, extract-spec-md, extract-imagen-prompts, extract-veo-prompts, extract-music-prompts, extract-pipeline-yaml, extract-input-text, extract-tree-view, extract-workflow
Run the command via Bash from the user's current working directory (never cd into the script directory). Use the full path to the script:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py <args>
The output .md file will be written to the user's current working directory. The script prints the output path to stderr. After execution, tell the user the output file path.
Important: Never output files to the skill's scripts directory. Always run from the user's cwd.
YouTube URL with default prompt:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py "https://www.youtube.com/watch?v=abc123"
Web page (blog post, article):
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py "https://www.latent.space/p/turbopuffer"
Local HTML file:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py --file "article.html"
Local PDF file:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py --file "paper.pdf"
Local transcript file:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py --file "/path/to/transcript.txt"
Custom inline prompt:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py "https://www.youtube.com/watch?v=abc123" -p "Extract all code snippets"
Built-in preset:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py "https://www.youtube.com/watch?v=abc123" --builtin takeaways
Transcript mode with speakers:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py "https://www.youtube.com/watch?v=abc123" --transcript --speakers "Alice" "Bob"
Segment analysis:
python ~/.claude/skills/yt-transcript-summary/scripts/yt_analyze.py "https://www.youtube.com/watch?v=abc123" --start 01:30 --end 05:00 -p "Summarize this segment"
GOOGLE_API_KEY must be set (in .env or environment)pip install -r requirements.txt (google-genai, python-dotenv, jinja2, requests, beautifulsoup4, pdfplumber)Launch a live dashboard showing all running Claude Code sessions on this machine. Use when the user asks to monitor sessions, check what's running, or view session activity.
360-degree Windows system health diagnostics — covers CPU, memory, disk, network, security, startup programs, services, system info, hardware, and installed software. Use this skill whenever the user mentions system diagnostics, PC health check, performance issues, 'why is my computer slow', disk space, memory usage, startup optimization, security audit, bloatware removal, system cleanup, laptop health, or wants to keep their PC in prime condition. Also use when the user asks about high CPU, RAM usage, what's eating their disk, or wants a full system report.
Convert raw YouTube transcripts into polished, high-fidelity readable articles in Markdown. Use this skill whenever the user wants to clean up, rewrite, or convert a YouTube transcript (or multiple transcripts) into an article, blog post, or readable document. Also triggers when the user provides transcript files (txt, html, pdf) or URLs and asks to make them readable, turn them into an article, or clean them up. Even if the user says something casual like "make this transcript readable" or "turn these into a blog post", this skill applies.
Create hand-drawn editorial illustration diagrams in Excalidraw with a sketchy, annotated, character-driven style. Use for AI workflow illustrations, technical comparisons, concept explanations, and visual storytelling.
End-of-day wrap-up that audits what Claude did automatically and what you did manually, then generates a beautiful HTML report with tomorrow's priority.
Generate a daily morning briefing dashboard that summarizes your calendar, email, and news.