Summarize Skill: One Command to Summarize Any Web Page, PDF, or YouTube Video
What This Skill Actually Does One sentence: Summarize lets you distill any content with a single command — web articles, PDF reports, YouTube videos, local a…
- Added to the lab
- Downloads
- 0
- Human review
- Human reviewed
What This Skill Actually Does
One sentence: Summarize lets you distill any content with a single command — web articles, PDF reports, YouTube videos, local audio files, all covered.
Real scenario: Your boss throws a 120-page industry report PDF at you and says "take a quick look." Before, that's two hours of reading. Now it's 15 seconds and you have the key takeaways. Or you subscribe to five English podcasts, each 90 minutes long, and simply can't keep up — Summarize extracts the key points so you don't have to sit through all of it.
With 211k+ downloads on ClawHub (ranked #2 overall, just behind self-improving-agent), this skill earned its popularity the hard way: it solves a real problem. Information overload is real, and Summarize cuts through it.
Installation and Setup
One command to install:
clawhub install summarizeAfter installation, the skill checks for the summarize CLI. If it's not there:
# macOS
brew install steipete/tap/summarize
# Or use npx (no install required)
npx summarize [URL or file path]Optional environment variables:
# Set default output language (optional)
export SUMMARIZE_LANG=en
# For audio transcription (optional)
export OPENAI_API_KEY=sk-xxxxxMost features work without an API key. You only need one if you're processing local audio files (mp3/wav) via Whisper. Web pages, PDFs, and YouTube videos work out of the box.
Real-World Usage
Not a Hello World — here's how this actually gets used.
Scenario 1: Summarize a competitor analysis report
# Feed it a PDF directly
summarize /path/to/competitor-analysis.pdf
# Or just say in OpenClaw:
# "Summarize the key points from this PDF" + attach fileScenario 2: Catch up on a YouTube talk
# Just the URL — no subtitle downloads needed
summarize https://www.youtube.com/watch?v=xxxxx
# Sample output:
# Key Points: 1. ... 2. ... 3. ...
# Key Data: ...
# Actionable Takeaways: ...Scenario 3: Batch summarize daily news
# Pass multiple URLs at once
summarize https://techcrunch.com/article1 https://example.com/article2
# Or pipe from a file
cat urls.txt | xargs summarizeScenario 4: Conversational use with OpenClaw
Once the skill is active, just say "summarize this article" and paste the URL in chat. OpenClaw automatically invokes the Summarize skill and responds conversationally. No manual CLI commands required.
Pitfalls and Gotchas
Issue 1: YouTube summarization fails
Error: Could not extract video info — usually means yt-dlp is outdated.
# Update yt-dlp
pip3 install --upgrade yt-dlp
# or
brew upgrade yt-dlpIssue 2: Garbled Chinese text in PDFs
Scanned PDFs (image-based) can't be text-extracted directly. Use the pdf skill's OCR function first, then pipe the output to Summarize.
Issue 3: Long content gets cut off
Default token limits truncate large files. Use --chunks for long documents:
summarize --chunks large-document.pdfIssue 4: Audio files need Whisper
Local audio (mp3/wav) requires OpenAI Whisper. Without an API key, install local Whisper: pip install openai-whisper. It's slower but free. SFD Lab runs local Whisper on Mac Studio — acceptable speed for most use cases.
Issue 5: Command not found after install
The skill installed but the CLI didn't. Make sure you ran brew install steipete/tap/summarize and that it's in your PATH.
How SFD Lab Uses It
Across our 13 agents at SFD Lab, Summarize shows up in three core workflows:
Content Agent (Fox🦊) + Summarize: Before writing any article, Fox uses Summarize to consume 3-5 reference pieces and extract key angles. Research that used to take 30 minutes now takes 5.
PM Agent (Raccoon🦝) + Summarize: Meeting recordings get fed into Summarize, which produces structured minutes and action items that go straight into the PRD. No manual transcription.
SEO Agent (Falcon🦅) + Summarize: Competitor content analysis — batch-summarize 20 top-ranking articles to find topic coverage gaps. Used to take three days. Now done in an hour.
We've also wired Summarize into our morning briefing pipeline: a cron job pulls 10 industry RSS feeds at 8am, Summarize batch-processes them, outputs a 100-word digest, and pushes it to Franky's Telegram. Stack: cron + summarize + message tool.
Pairing With Other Skills
Summarize + deep-research-pro: deep-research-pro gathers sources, Summarize digests each one. Together they run a full "research → extract → report" pipeline with zero human intervention.
Summarize + voice-message: Summarized content goes straight to audio. SFD's daily briefing works exactly like this — Summarize generates text, edge-tts converts to audio, and it lands in Telegram.
Summarize + smart-web-scraper: Scraper crawls a batch of pages, Summarize processes each one. Great for monitoring competitor site updates — new content triggers automatic summarization and a notification.
Summarize + multi-agent-cn: In a multi-agent architecture, Summarize acts as the information preprocessing layer. Downstream agents receive structured summaries instead of raw long-form content, cutting token usage significantly.
SFD Editor's Note: Summarize is one of our must-have skills at SFD Lab. Being ranked #2 on ClawHub isn't an accident — it solves the most practical problem of the AI era: too much information, not enough time to process it. If you're consuming large amounts of content daily, this skill can double your efficiency overnight. Install it, use it, no learning curve required. — XiaoHuolong🔥