alphaxiv-paper-lookup

name	alphaxiv-paper-lookup
description	Look up arXiv papers through alphaXiv and return a structured AI-generated overview plus the arXiv abstract. Use when the user provides an arXiv URL, alphaXiv URL, or paper id; asks to summarize/explain/analyze a paper; or wants a faster overview before reading the PDF.

alphaXiv Paper Lookup

Use this skill to turn an arXiv identifier or URL into a fast, structured paper brief. Prefer alphaXiv first because it often exposes an AI-generated overview that is faster to scan than the raw PDF. Fall back to arXiv when alphaXiv is missing, thin, or unavailable.

Quick workflow

1. Normalize the input into one or more paper ids.
   • Accept direct positional arguments and optional file-driven inputs.
   • Accept plain ids like 2401.12345 or 1706.03762v7
   • Accept arXiv URLs like https://arxiv.org/abs/2401.12345
   • Accept alphaXiv URLs like https://www.alphaxiv.org/overview/2401.12345
2. Run the bundled script:
   • The script accepts one or more paper ids / URLs in a single invocation.
   • Use --input-file PATH to add repo-local batch inputs.
   • Plain-text inputs stay line-based: read one id / URL per line, ignoring blank lines and lines starting with #.
   • CSV/TSV inputs use a header row. Prefer --column COLUMN_NAME to select the input column explicitly.
   • If --column is omitted for CSV/TSV, the script only auto-selects an obvious single input column; otherwise it fails and prints the available columns.
   • python3 scripts/alphaxiv_lookup.py "<paper-or-url>" --format markdown
   • python3 scripts/alphaxiv_lookup.py --input-file papers.csv --column paper_id --format json
   • Use --format json for full structured output.
   • Use --format json-compact when you want a smaller machine-friendly payload.
   • Use --format text for a clean plain-text brief.
   • Use --format brief for a concise user-ready brief.
   • Use --format brief-zh for the same structure with Chinese labels.
   • For batch usage: json returns {count, results}, json-compact returns JSONL, and human-readable formats emit multiple blocks with separators.
3. Read the returned fields in this priority order:
   • best_summary
   • alphaxiv_report
   • alphaxiv_description
   • arxiv_abstract
   • source_used, summary_source, status, alphaxiv_status, arxiv_status, warnings, errors
4. Write the answer in a fixed structure:
   • Paper title
   • What problem it solves
   • Core idea / method
   • Key findings
   • Limitations / caveats
   • Whether it is worth reading in full
5. If the user asks for deeper analysis, use the extracted report + abstract to produce:
   • Method breakdown
   • Comparison to prior work
   • Practical implications
   • Open questions

Output template

Use this template unless the user requests a different style:

• Paper: title + id
• 一句话结论: what the paper claims in one sentence
• 解决什么问题: task / pain point / gap
• 核心方法: 3-5 bullets
• 关键结果: benchmarks, ablations, or empirical takeaways
• 局限性: assumptions, missing comparisons, scalability, data dependence
• 值不值得细读:
  • 值得细读 / 值得略读 / 只看摘要即可
  • give one short reason

Fallback rules

• If alphaXiv returns only a thin description, combine it with the arXiv abstract instead of pretending the overview is complete.
• If alphaXiv fetch fails, classify it as precisely as possible (rate_limited, not_found, upstream_error, or generic failure) and fall back to arXiv.
• If the user asks for exact equations, implementation details, or appendix-level nuance, warn that alphaXiv is only a shortcut and the PDF/source paper should still be checked.
• Do not invent missing benchmark numbers. If a metric is absent, say it is not surfaced in the retrieved overview.

Good trigger examples

• “帮我总结这篇 arXiv：2401.12345”
• “看看这个论文讲了什么 https://arxiv.org/abs/2401.12345”
• “解释一下这篇 paper 的核心贡献”
• “这个 alphaXiv 链接值不值得读原文？”
• “先帮我快速过一遍这篇论文，再决定要不要看 PDF”

Resource

scripts/alphaxiv_lookup.py

Normalize input, fetch alphaXiv overview pages, extract embedded report/description fields when available, and fall back to the arXiv abstract.