AgentSkillsCN

uv-read-arxiv-paper

下载并深入研读 arXiv 论文（给定 arXiv URL 或 ID），随后撰写清晰易懂、兼具故事性和逻辑性的面向人类的报告。当您被要求总结/评审 arXiv 论文、提炼关键思想、将其与实际应用相联系，并在 $HUMAN_MATERIAL_PATH/research/<paper_slug>/report.md 下生成报告时，可运用此技能。下载的文献将存储于 $HUMAN_MATERIAL_PATH/.references/（可通过 $HUMAN_MATERIAL_PATH/.agents/config.toml 或 ~/.agents/config.toml 进行配置）。

SKILL.md
--- frontmatter
name: uv-read-arxiv-paper
description: "Download and deeply read an arXiv paper (given an arXiv URL or id), then write a clear human-facing report with strong storytelling and logical reasoning. Use when asked to summarize/review an arXiv paper, extract key ideas, connect them to practice, and produce a report under $HUMAN_MATERIAL_PATH/research/<paper_slug>/report.md. Stores downloads under $HUMAN_MATERIAL_PATH/.references/ (configurable via $HUMAN_MATERIAL_PATH/.agents/config.toml or ~/.agents/config.toml)."

Read an arXiv paper (deep report)

Output locations (pkbllm convention)

  • •Tracked output (write here): $HUMAN_MATERIAL_PATH/research/<paper_slug>/
    • •report.md (main deliverable)
    • •figures/ (optional images/diagrams)
  • •Local downloads (gitignored, configurable): $HUMAN_MATERIAL_PATH/.references/
    • •PDFs default: .references/pdfs/
    • •Sources default: .references/arxiv/
    • •Cloned repos default: .references/repos/

Config precedence (repo overrides user):

  1. •$HUMAN_MATERIAL_PATH/.agents/config.toml
  2. •~/.agents/config.toml

Quick start

  1. •Download paper assets:
bash
python skills/uv-read-arxiv-paper/scripts/download_arxiv.py https://arxiv.org/abs/2601.07372
  1. •Extract the TeX source:
bash
python skills/uv-read-arxiv-paper/scripts/extract_arxiv_source.py 2601.07372
  1. •Write the report:
  • •Create $HUMAN_MATERIAL_PATH/research/<paper_slug>/
  • •Start from skills/uv-read-arxiv-paper/assets/report_template.md
  • •Fill it with a coherent story:
    • •What problem is being solved and why now?
    • •What is the key trick (the one idea that makes it work)?
    • •What are the assumptions and failure modes?
    • •What to implement first if we were to reproduce it?

Workflow (recommended)

  1. •Normalize the paper id from the URL (keep the version if present; it’s fine).
  2. •Download PDF + source into .references/ using the provided scripts.
  3. •Unpack sources into .references/arxiv/<arxiv_id>/ and locate the LaTeX entrypoint (main.tex, etc.).
  4. •Read in passes:
    • •Pass 1: abstract + intro + figures (build the mental model)
    • •Pass 2: method (write down the algorithm precisely)
    • •Pass 3: experiments (what matters, what doesn’t)
    • •Pass 4: limitations + related work (what breaks, what’s missing)
  5. •Produce a report under $HUMAN_MATERIAL_PATH/research/<paper_slug>/report.md:
    • •Use the template (copy it and fill it)
    • •Prefer concrete examples, small equations, and “why this design” explanations
    • •Include at least one visual explanation if it helps (optionally generate via uv-scientific-schematics)