Defuddle: fetch and clean a web page

Defuddle is a fetch-and-clean step, not a filing step. Cleaning a page and filing it as a note are separate operations, each requiring its own consent — this skill does the first and, only when asked, hands off to the same note-writing rules capture uses for the second.

Treat Defuddle as an optional external extractor the host environment provides — this skill never installs one, executes a placeholder, or silently substitutes another fetcher.

Safety contract

• Accept remote input only as an HTTPS URL.
• Reject: credentials embedded in the URL, URL fragments used as a smuggling vector, private/local hosts (localhost, loopback, link-local, RFC1918 ranges, etc.), non-public IP literals, control characters, and sensitive query parameters (tokens, keys, session ids).
• Never interpolate the URL into a shell string — pass it as a single argv element to the extractor.
• Treat a redirect to a different host as denied until that host is explicitly approved.
• State plainly that the URL and request metadata will leave the machine. Do not fetch without explicit consent in the current request (or a separate confirmation the user already gave for this URL).
• Do not claim a fixed token reduction or extraction quality — inspect the actual output.

Plan before fetching

Validate the URL against the safety contract above yourself — there's no script to run this for you. Report the normalized host, what leaves the machine, the redirect policy, and that no network call has happened yet.

Check whether an external Defuddle-style extractor is available (e.g. command -v defuddle, or ask the user). Three outcomes:

• Not found: stop here. Offer honest fallbacks: the user installs/configures one, pastes the page content directly for capture to use instead, or leaves the URL for later.
• Found, unreviewed: this skill has no bundled verifier for a discovered executable. Show its resolved path and require the user to confirm its provenance and version before the first use — never silently trust a command -v hit.
• Found and already reviewed this session: continue.

Fetch after consent

Invoke the reviewed executable with an argv equivalent to:

defuddle parse HTTPS_URL --md

Fail closed on: non-zero exit, empty output, unexpected binary output, an unapproved redirect, or a response that's clearly an auth/error page. Preserve headings, links, code fences, tables, and quotations exactly — never invent missing content.

Preview the cleaned Markdown and report extraction limitations. If the user only wanted to read or analyze it, stop here — the result stays transient, nothing is written.

File as a note (optional, separate consent)

If the user wants the cleaned page kept, file it the same way capture would — read the note format and vault layout reference and follow its rules:

1. Derive a title and slug from the page; check the whole vault for a filename collision before writing (never silently overwrite).
2. Decide the destination — projects/<slug>/, notes/<topic>/, or inbox/ — per folder placement, checking indexes/ before creating a new notes/ topic folder and asking the user if it's genuinely ambiguous between project and general note.
3. Write the note using the current template's roles: source (the HTTPS URL), content holding the cleaned page's key points or a short excerpt — not the full cleaned page verbatim unless it's already short — and connections if anything is genuinely related. Keep the original cleaned Markdown available in the conversation in case the user wants more of it; don't dump the entire page into the vault by default.
4. If the note landed in notes/<topic>/, update the index for that topic.
5. Report the note's path, the index updated (if any), and any links added.

No transaction bundle, no ledger, no .raw/ payload store. This skill never marks anything as canonical or ingested on its own.