Wikipedia Skill Pack Files

Wikipedia

Article extraction, content reading, and summarization. Agents read the public article namespace, the talk pages, and category indexes. No login needed for reads; editing is out of scope for this pack.

Browser Access

No login required. Wikipedia is fully public and uncached behind authentication. The same selectors work across language editions because all Wikimedia wikis share the MediaWiki theme — only the URL host changes (en.wikipedia.org, de.wikipedia.org, fr.wikipedia.org, etc.).

Page types

URL pattern	Purpose
/wiki/<Title>	Article (default namespace)
/wiki/Talk:<Title>	Discussion page for the article
/wiki/Category:<Name>	Category index — lists member pages
/wiki/File:<Name>	Image / media metadata page
/wiki/Special:<Page>	Auto-generated tools (Random, RecentChanges, Search)
/w/index.php?title=<Title>&action=history	Revision history
/w/index.php?title=<Title>&action=edit	Wiki source view (read-only safe)

Article structure

Every article uses the same anatomy:

Section	Selector	Notes
Title	#firstHeading	Canonical page title, H1
Lead paragraph	First <p> inside #mw-content-text	Best single-paragraph summary
Infobox	.infobox	Right-rail key/value table — birth dates, founding years, taxonomy
Table of contents	#toc	Anchor navigation built from ## / ### headings
Body	#mw-content-text > .mw-parser-output	All section bodies
References	.references, ol.references li	Numbered inline citation list
See also	#See_also heading section	Editor-curated related links
External links	#External_links heading section	Outbound primary sources

Infoboxes are the most data-rich part of the page and parse cleanly into key/value pairs. They are the right target when extracting structured facts (population, area, capital, founder, etc.).

Disambiguation

When a title resolves to a disambiguation page, MediaWiki adds <div class="hatnote"> at the top and the body becomes a list of links. Detect this with the page categories: disambiguation pages carry Category:Disambiguation_pages. When found, fall back to a more specific search query rather than extracting the page body as if it were an article.

Citations and references

Inline citations render as superscript anchor links (<sup id="cite_ref-…">[1]</sup>) that point to entries in .references. Each reference list item contains the citation text plus an outbound link to the source. To extract a citation graph for an article, walk the references list and resolve each <a> href.

Categories

The footer of every article shows its categories at #mw-normal-catlinks. Categories form a graph: each is itself a /wiki/Category:<Name> page that lists members. Walking categories breadth-first is the standard way to enumerate "all articles about X" without using the search index.

Common tasks

Summarize article: navigate to /wiki/<Title>, extract the lead paragraph (first <p> inside #mw-content-text) and the first sentence of each top-level section, hand to an LLM for synthesis.

Extract content: navigate to article URL, snapshot #mw-content-text, optionally drop .reference, .mw-editsection, .thumb, and .navbox to clean prose.

Extract infobox: select .infobox tr, parse th (label) + td (value) pairs.

List a category: navigate to /wiki/Category:<Name>, walk pagination through next page links to enumerate all members.

Gotchas

• Some pages use #mw-content-text but the parsed output is wrapped in .mw-parser-output — both selectors are needed for a robust grab.
• Section anchors are URL-encoded versions of the heading text with spaces as underscores (#See_also).
• The mobile site (en.m.wikipedia.org) has a different DOM and collapses sections by default — prefer the desktop host.
• Infoboxes vary by article type (Person, Place, Company, Film, Album, …) — keys are not standardized across types.
• Wikipedia rate-limits aggressive crawling. Between articles add a 1–2 second pause and respect any Retry-After headers if scraping at volume.