Extract the readable content of a web page with defuddle, a CLI that parses HTML from a URL, file, or stdin and returns the main article content with its metadata — navigation, ads, and boilerplate stripped. Where curl gives you a raw HTML dump, defuddle gives you text worth reading: Markdown tables, preserved code fences, and structured metadata such as title, author, publish date, and word count.

Use when:

• “Extract this article”
• “Convert this page to markdown”
• “Get the main content of this URL”
• “Clean up this HTML”
• “Save a readable snapshot of this link”
• Any request for a page’s readable text rather than its raw HTML

What it does

• Parses a URL, a local HTML file, or stdin (defuddle parse -)
• Outputs Markdown (--markdown), full metadata JSON (--json), or a single field (--property wordCount — camelCase, matching the JSON keys)
• Prepends YAML frontmatter (title, site, source, domain, language, word_count) with --frontmatter; combined with --markdown --output <file> this writes a complete page snapshot in one step
• Retries blocked fetches with a custom user agent (--user-agent); accepts a language hint with --lang

Usage

Invoke the skill by name, or let a supporting agent select it for a relevant request. Automatic activation depends on the host agent and its configuration.

Examples:

Extract the main content of https://example.com/article as markdown

Save a readable snapshot of this link with frontmatter

What's the publish date and author of this article?

The skill treats each extraction as a claim to verify: fetch errors print but exit 0, so the output is checked directly; empty bodies, login walls, and paywall stubs are reported as failed extractions instead of being summarized as the source. JavaScript-rendered pages that return little content are named as a limitation, with a fallback attempted before giving up.

Defuddle is an external dependency (npm package defuddle). Agents run it via npx --yes defuddle@<version> with a pinned version when reproducibility matters, and record the version used when an extraction is kept. When defuddle feeds a wiki-ingest URL snapshot, wiki-ingest remains in charge of provenance, retrieval date, and commit discipline.

Full Skill Definition

See skills/defuddle/SKILL.md for the complete agent instructions — output-format selection, the verified property list, extraction checks, and boundaries — that the agent loads when this skill is active.