Summarize

You can summarize content from various sources: web pages, articles, YouTube videos, and files.

Summarizing a web page

1. •Fetch the page content:

# Method 1: curl + HTML strip (works everywhere)
curl -sL "URL" | sed 's/<script[^>]*>.*<\/script>//g' | sed 's/<style[^>]*>.*<\/style>//g' | sed 's/<[^>]*>//g' | sed '/^[[:space:]]*$/d' | head -500

# Method 2: readability-cli (better extraction, if installed)
readable "URL" 2>/dev/null | head -500

1. •Then summarize the extracted text using your reasoning capabilities.

YouTube video transcripts

# Get auto-generated subtitles (requires yt-dlp)
yt-dlp --write-auto-subs --skip-download --sub-lang pt,en -o "/tmp/%(id)s" "VIDEO_URL" 2>/dev/null
cat /tmp/*.vtt 2>/dev/null | grep -v "^[0-9]" | grep -v "^$" | grep -v "WEBVTT" | grep -v "^NOTE" | grep -v "\-\->" | sort -u

# Alternative: get video info first
yt-dlp --print title --print duration_string --print description "VIDEO_URL" 2>/dev/null

PDF files

# Extract text from PDF (requires pdftotext)
pdftotext /path/to/file.pdf - | head -500

# Alternative with python
python3 -c "
import subprocess
text = subprocess.check_output(['pdftotext', '/path/to/file.pdf', '-']).decode()
print(text[:5000])
"

Summarization guidelines

When summarizing, follow these best practices:

1. •Brief summary (default): 3-5 bullet points with the key takeaways.
2. •Detailed summary: Structured with sections, preserving important details.
3. •TL;DR: One sentence capturing the essence.

Always:

• •Preserve key facts, names, dates, and numbers.
• •Maintain the original language unless the user asks otherwise.
• •For technical content, keep important code snippets and terminology.
• •Mention the source URL/title in your summary.
• •Ask the user what level of detail they want if not specified.

Tips

• •Combine with web-search to find content, then summarize it.
• •For very long content, break into sections and summarize each.
• •Use head -N or tail -N to limit input when content is too large.
• •If yt-dlp is not installed: pip install yt-dlp or brew install yt-dlp.

Triggers

summarize, summary, transcribe, transcript, what does this video say, read and summarize, tl;dr, resumir, resumo, transcrever