Analyze academic papers and generate beautiful interactive HTML analysis pages. Use this skill whenever the user wants to: analyze a paper, review a paper, understand a paper's contributions, create a paper summary, generate a paper analysis report, or when they share an arXiv URL or PDF and ask for analysis. Also trigger when the user mentions "paper analysis HTML", "paper review page", or asks to create a visual/interactive summary of a research paper. This skill produces a dark-themed, responsive HTML page with 7 structured sections, each containing original quotes and intuitive explanations with analogies.
Generate interactive, dark-themed HTML pages that deeply analyze academic papers with a consistent 7-section structure. Each section pairs original quotes from the paper with intuitive explanations and analogies accessible to non-experts.
Before doing anything else, read references/example-output.html in full. This is the gold-standard reference. Your generated page must match its tone, depth, and structural granularity. Specifically observe:
linear-gradient(135deg, #1e3a5f 0%, #0f172a 50%, #1a1040 100%)) — never invent a new one<em> tags on key terms<strong>核心概念:</strong> (or <strong>Key idea:</strong> for English), then 2-4 short paragraphsYour job is to match this exemplar's style for any new paper, not to re-invent the look each time.
Try these approaches in order:
https://arxiv.org/html/<id> or https://arxiv.org/html/<id>v<N> — use WebFetchhttps://arxiv.org/abs/<id> — use WebFetch for metadatahttps://arxiv.org/pdf/<id> — use WebFetch to download, then Read the PDFpages parameterFor PDFs, read in batches (pages 1-10, then 11-20 if needed) to capture full content.
From the paper, extract and organize:
For each point, collect:
Generate the HTML file by combining two references:
references/example-output.html — the style, tone, and structure you must matchreferences/template-guide.md — the component specs (CSS classes, HTML snippets, required blocks)The exemplar is the primary reference; the template guide is the technical spec. If the two ever seem to conflict, follow the exemplar.
Output location: Save to the user's working directory as paper_analysis_<SHORT_ID>.html where <SHORT_ID> is a recognizable identifier (e.g., WSDM25, FCS24, NeurIPS23_attention).
Tell the user the file is ready and visible in the preview panel. Mention the 7 sections and offer to explain any part in more detail.
<em> tags — not moreDetect the user's language from the request message using this exact rule:
<html lang="zh-CN">, use Chinese section headers and Chinese labels (📄 原文, 💡 通俗解释, 🎯 打个比方)<html lang="en">, use English section headers and labels (📄 Original, 💡 Intuitive Explanation, 🎯 Analogy)Paper quotes always stay in the paper's original language regardless of UI language.
Every generated page must have these 7 sections:
| # | Section | Color Accent | What to Include |
|---|---|---|---|
| 1 | Research Background | Blue | Problem definition, why it matters, key statistics |
| 2 | Existing Methods | Purple | 3-5 method categories with brief explanations |
| 3 | Limitations of Existing Methods | Orange | 2-4 specific limitations, with summary table |
| 4 | This Paper's Innovations | Green | 2-4 innovation points, each with quote+explain+analogy |
| 5 | Method Details | Blue | Core formulas, workflow diagram (using flow components), key design choices |
| 6 | Conclusions & Findings | Pink | 3-5 key findings from experiments |
| 7 | Limitations & Future Work | Red | Paper's own stated limitations + potential directions |
Each section uses: section-icon with number → cards containing quote blocks, explain blocks, and optionally analogy blocks, tables, or stats grids.
When the paper fits into a series or relates to other papers the user mentioned, add a comparison table in Section 7 showing how papers relate to each other (different perspectives, methods, scope).
Before delivering, verify: