← ClaudeAtlas

compare-paperslisted

Distil a set of 2–20 papers into a tight 4–6 sentence comparison: what the set agrees on, the most important divergence or conflict, and which paper to read first and why. Speaks about the set as a whole, not paper-by-paper. Use when a user drops several papers and asks "how do these compare", "what's the consensus / disagreement here", "which should I read first", or wants a quick cross-paper take rather than a full literature overview.
jy1529098645-gif/VsCat · ★ 0 · AI & Automation · score 70
Install: claude install-skill jy1529098645-gif/VsCat
# Compare Papers The fast cross-paper read. Given 2–20 papers, produce a short, dense comparison that a researcher can absorb in twenty seconds — the opposite of a per-paper summary. Use this when someone has a few papers open and wants to know *how they relate*, not what each one says. (For a longer, structured synthesis with sectioned analysis and inline source markers, use the `literature-overview` skill instead. This one is deliberately brief.) ## Input 2 to 20 papers. For each: title, authors, year, and abstract/summary. Trim long abstracts — you only need enough to locate each paper's stance. ## Prompt ``` You are an academic literature reviewer. Read the paper set below and produce a tight 4-6 sentence comparison in plain English. Cover: (1) the central agreement across the set, (2) the most important divergence or conflict, and (3) which paper a reader should prioritise and why. Do NOT enumerate papers individually; speak about the set as a whole. Avoid boilerplate openers like "This set of papers". Output prose only — no headings, no bullet points. Paper 1: {title} by {authors}, {year} abstract: {summary} Paper 2: ... ``` ## Rules - **4–6 sentences. Prose only** — no headings, no bullets. - **Speak about the set as a whole.** Never "Paper 1 says… Paper 2 says…". The value is in the relationships: where they converge, where they collide. - Hit all three beats: **agreement → key divergence → what to read first (and why).** - **Weight by evidenc