web-fetch (tonyfettes/skills-web-fetch)

• DeepSeek Fetch — POST https://api.deepseek.com/v1/tools/fetch returns the page already rendered as Markdown, with DeepSeek's own extractor doing the work. Needs an API key and is billed per successful request.
• Local — pure-MoonBit async HTTP plus the WHATWG moonbit-community/html parser: redirects are followed, relative links resolved, HTML reduced to its title and main text. No key, no cost, no third party.

Use This Skill For

• Reading the content of a documentation page, blog post, changelog, wiki, GitHub page, or any server-rendered site (SSR output parses fine).
• Checking a page's title/headings or pulling the absolute links it contains.
• Reading JSON or plain-text endpoints that a page or API returns — with --backend local, which returns the bytes unchanged.

• The page is a client-side-only app (React/Vue shell) — the fetched HTML has no content because MoonBit cannot execute the page's JavaScript. There is no CSR-detection heuristic in this tool, so it will happily print what the server actually sent: an empty shell. Recognize that situation yourself and fall back to the site's JSON API, or say a headless browser would be needed. The DeepSeek backend is sometimes better here, and sometimes not: whether its extractor runs a browser is not documented, so treat a thin result the same way and check.
• You need a screenshot or full DOM after scripts run.

Running (published package)

moonx tonyfettes/skills-web-fetch <url>
moonx tonyfettes/skills-web-fetch --links <url>
moonx tonyfettes/skills-web-fetch --backend local --raw <url>
moonx tonyfettes/skills-web-fetch --backend deepseek --text 20000 <url>
moonx tonyfettes/skills-web-fetch --help

moon run --target native . -- <url>
moon test                       # unit tests across internal/* packages
moon cram test tests/cram       # offline CLI contract tests
moon build --target wasm        # what moonx will run

Named script (SeekMoon / OpenSeek)

# <skill-dir> is the directory this skill is listed under, e.g.
#   /Users/you/.openseek/skills/tonyfettes-skills-web-fetch       (macOS)
#   /home/you/.openseek/skills/tonyfettes-skills-web-fetch        (Linux)
#   C:\Users\you\.openseek\skills\tonyfettes-skills-web-fetch     (Windows)
mbtx(filename="<skill-dir>/scripts/fetch.mbtx", args=["<url>"])

Options

Backends

Output Shape

Title: Example Domain

# Example Domain

This domain is for use in documentation examples without needing permission. Avoid use in operations.

• The Title: line and the blank line after it are omitted when the backend found no title. The body follows as Markdown (#–###### headings, - list items), truncated at --text chars with a trailing [...] marker.
• Locally, extraction starts at the first <main> or <article>; otherwise the whole <body> is used, so site chrome (<nav>, menus) is usually excluded. <script>, <style>, <template>, SVG, and friends are dropped; <pre> text keeps its formatting. It is a lightweight rendering, not a faithful HTML→Markdown conversion.
• Local non-HTML responses (JSON, text/plain, XML, or an HTML-looking body without an HTML content type) print their body as-is.
• --links works on both: locally the links come from the parsed DOM, on the DeepSeek backend from the [label](url) pairs in the returned Markdown. Relative targets are skipped there, and an empty label falls back to the URL.

Errors and exit codes

• 0 — every requested page printed.
• 1 — at least one page failed; each failure prints one error: ... line, the rest are still attempted.
• 2 — usage error (bad option or value, no URL given, --backend deepseek without a key, --raw together with --backend deepseek).

error: DeepSeek Fetch HTTP 404 for <url>: could not retrieve usable page content
error: DeepSeek Fetch HTTP 401 for <url>: invalid or missing API key: <provider message>
error: could not reach the DeepSeek Fetch API (<base-url>) for <url>: network or TLS failure
error: URL must contain a host for <url>
error: URL must not contain a username or password for <url>

Practical guidance

• Keep --text small when the page might be long and you only need an overview; raise it when you need full prose.
• Use --links to find the next hop instead of hand-parsing HTML.
• Use --raw only when you need the exact markup; it is local-only.
• Locally, bodies are decoded as UTF-8 only; legacy charsets (e.g. GBK, latin-1) come out mojibake — say so rather than pretending the text is accurate. The DeepSeek backend decodes and renders server-side, so it may do better or worse; if the text looks wrong, try the other backend.
• Redirects are followed locally up to 10 hops (Location header resolution included); the DeepSeek backend follows them on its side.
• The local backend is anonymous: no cookies or ambient credentials, and no SSRF filtering — do not point it at private/internal URLs or anything that requires auth. The DeepSeek backend applies its own security policy and returns 403/404 instead.