use-defuddle

Essential Principles

• Always use --md - markdown output is more useful than raw HTML for note-taking, summarisation, and storage. Only omit for HTML-specific processing
• Use --json for metadata - when you need title, author, date, description, or domain alongside the content. JSON output includes both metadata and content in a single call
• Use -p for single properties - defuddle parse <url> -p title is cheaper than --json when you only need one field
• Prefer defuddle for articles and documentation - it strips navigation, ads, sidebars, and comments. Use the harness's raw web fetch/extract tool for raw Markdown URLs or when you need the full unprocessed page
• Confirm with user before writing files - defuddle -o writes to disk; derive filename from -p title when not specified, then confirm before writing
• Fall back to the available web fetch/extract tool on empty output - paywalled pages, JavaScript-heavy SPAs, and some dynamic sites produce empty or incomplete content. Use web_extract in Pi, WebFetch in Claude Code, or ask the user for pasted/exported content if no web fallback exists

Prerequisites

command -v defuddle >/dev/null 2>&1

If not found: npm install -g defuddle.