| name | wiki_dive |
| description | Deep Wikipedia research — go beyond the keyword search by reading articles section-by-section, following related links, and synthesising a structured report with citations. Use when the user wants a deep dive, primer, or thorough background on a topic. |
| requirements | [] |
| examples | ["Deep dive on the Cambrian explosion","Tell me everything Wikipedia says about transformer (deep learning)","Background on the Apollo program","What's the encyclopedia view on cellular automata?"] |
Wiki Dive — Deep Wikipedia Research
You help users understand complex topics by reading Wikipedia articles
thoroughly — not just the lead summary, but full sections, cross-links,
and related articles. A companion script — scripts/wiki_tools.py —
exposes four CLI subcommands.
When to use this skill
Trigger on any request that involves:
- "Deep dive / primer / thorough background on <topic>"
- "Tell me about <X>" (encyclopedic intent)
- "What does Wikipedia say about <Y>"
- "Compare <A> and <B>" with an encyclopedic, neutral framing
Don't use this for current news, opinion, or product reviews — Wikipedia
isn't the right source.
Tools provided
| Subcommand | Purpose | Returns |
|---|
search_wikipedia <query> [max_results=6] | Find relevant article titles by keyword. | {"results": [{title, snippet, url}, ...]} |
get_article_summary <title> | Lead summary (a few paragraphs) of a named article. | {"title", "summary", "url", "thumbnail"} |
get_article_sections <title> | Full plain-text article via the action API. | {"title", "extract", "url"} |
get_related_articles <title> [max_results=8] | Internal links from the article — discover related concepts. | {"source", "related": [{title, url}, ...]} |
Pass titles exactly as Wikipedia uses them (case-sensitive, spaces
allowed). If the title isn't known, search first.
Example invocation
python scripts/wiki_tools.py search_wikipedia 'Cambrian explosion'
python scripts/wiki_tools.py get_article_summary 'Cambrian explosion'
python scripts/wiki_tools.py get_article_sections 'Cambrian explosion'
python scripts/wiki_tools.py get_related_articles 'Cambrian explosion' 8
Workflow
Topic research
search_wikipedia(query) to find the most relevant article(s).
get_article_summary(top_title) on the top 1-2 hits to confirm
relevance. Discard disambiguation pages — try a refined title.
get_article_sections(primary_title) for deep content. You must
call this — summarising the lead alone is not a deep dive.
get_related_articles(primary_title) to discover connected concepts.
get_article_summary on 2-3 related articles that add meaningful
context (predecessor concepts, competing theories, key figures).
- Synthesise across all articles in the format below.
Direct article request
If the user names an article ("the Wikipedia article on X"), skip the
search step and go straight to get_article_sections(title).
Citation format
Every claim from Wikipedia MUST cite its source article inline:
According to Article Title: "key fact or close paraphrase"
When multiple articles confirm a point:
"Both Transformer (deep learning) and Attention
mechanism (machine learning) describe self-attention as …"
Output structure
**Topic**: <topic>
**Articles read**
- [Title](url) — one-line description
- ...
**Overview** (2-3 paragraphs)
<plain-language synthesis. No jargon without explanation. Cite inline.>
**Key concepts**
- <concept> — 1-2 sentences with source article cited
- ...
**History / development** (if relevant)
<chronological narrative with citations>
**Current state / applications** (if relevant)
<what is this used for today; where is it going>
**Points of debate or nuance** (if any)
<contested views, ongoing research, or limitations Wikipedia notes>
**Related topics to explore**
3-5 linked concepts with one-line descriptions and Wikipedia URLs.
Tone & failure modes
- Encyclopedic, neutral tone — match Wikipedia's style.
- Never fabricate facts. Report only what the tools return.
- If an article doesn't exist or is a disambiguation page, try a refined
title. If still empty, say so plainly.
- If Wikipedia coverage is sparse, say so and explain the gap.
- Keep the synthesis under 800 words unless the user asks for more
depth.
- If your host has no way to execute the script (no shell or subprocess
primitive), say so plainly. Do not guess at article content.