Survey the latest follow-up work for a target paper by first retrieving recent citations, listing the newest 20 cited papers for user selection, then reading the selected papers and summarizing their directions and how they extend the original paper. Use when the user asks things like "调研一下某篇 paper 的最新工作", "看看这篇论文最近有哪些 follow-up", "列出这篇论文最新 citations 并帮我读几篇", or wants a citation-driven literature update around a seed paper.
Use this skill to turn a seed paper into a citation-driven mini survey. Always first gather recent citing papers, then let the user choose what to read, and only after selection analyze the chosen papers in relation to the seed paper.
Extract the seed paper identifier from the user request.
Accept arXiv URL, arXiv id, DOI, or Semantic Scholar id.
If the paper is an arXiv paper, normalize it to the bare arXiv id for reading and to ARXIV:{id} for citation lookup.
Invoke the paper-citations skill first.
Use it to fetch citation papers for the seed paper.
Prefer the newest papers by publication date when available; otherwise sort by year descending and keep recent results first.
If no citations are found, tell the user there are no indexed citations yet and stop.
From the citation list, produce a shortlist of at most 20 newest papers. For each paper, include:
Then ask the user which papers they want to read. Support selection by:
Do not read papers before the user selects them unless the user explicitly asks for automatic reading.
For every selected arXiv paper, invoke the arxiv-paper-reading skill.
If a selected citation is not on arXiv, say that this workflow currently reads arXiv papers directly and ask whether to skip it or replace it with another shortlisted paper.
When multiple selected papers are available, read them one by one and keep notes on:
After reading the selected papers, summarize them in two layers.
First give a per-paper summary using this compact structure:
Then provide a cross-paper synthesis:
Always distinguish paper-supported claims from your own inference when the relation is not explicitly stated by the citing paper.
When showing the shortlist, use a concise numbered list. When summarizing selected papers, keep each paper to a short block, then end with a grouped trend summary.
Do not fabricate citation metadata. Do not assume every citation is truly technically relevant; some may be only loosely connected. If the citation list contains obviously unrelated papers, mention that relevance is based on citation indexing plus title-level inspection unless the paper was actually read. If publication date and year disagree, prefer the more specific publication date.