When to Use
User wants to compare multiple papers and may provide paper titles, links, or PDF files.
Use this skill when the goal is to identify differences, strengths, weaknesses, assumptions, and trade-offs across papers.
Quick Reference
| Topic | File |
|---|
| Input handling | INLINECODE0 |
| Comparison schema |
comparison-schema.md |
| Output formats |
output-patterns.md |
| Evidence rules |
references/evidence-policy.md |
Core Rules
- 1. Retrieve and read PDF full text for every paper before doing substantive comparison. Do not rely on abstract-only evidence.
- Normalize all inputs into comparable paper records before comparing them.
- Compare papers on shared dimensions only, and avoid direct metric comparison when evaluation settings differ.
- Mark evidence quality and uncertainty explicitly for each paper and each major claim.
- Produce a table-first comparison, then add concise analytical synthesis.
- If a full PDF cannot be obtained for one or more papers, stop the comparison and report which papers are blocked.
何时使用
用户希望比较多篇论文，可能提供论文标题、链接或PDF文件。
当目标是识别不同论文之间的差异、优势、劣势、假设和权衡时，使用此技能。
快速参考
| 主题 | 文件 |
|---|
| 输入处理 | input-handling.md |
| 比较模式 |
comparison-schema.md |
| 输出格式 | output-patterns.md |
| 证据规则 | references/evidence-policy.md |
核心规则
- 1. 在进行实质性比较之前，检索并阅读每篇论文的PDF全文。不得仅依赖摘要作为证据。
- 在比较之前，将所有输入标准化为可比较的论文记录。
- 仅基于共同维度比较论文，当评估设置不同时，避免直接比较指标。
- 对每篇论文和每个主要主张，明确标注证据质量和不确定性。
- 先以表格形式进行比较，然后添加简洁的分析性总结。
- 如果无法获取一篇或多篇论文的完整PDF，则停止比较，并报告哪些论文被阻止。
