| name | paper-comparison |
| description | Compare two or more note cards, DOI/arXiv/OpenAlex identifiers, direct PDF URLs, uploaded PDFs, or an explicitly scoped topic-plus-count request on a user-defined research question, method, evidence, limitation, and contribution axis. |
| tools | ["read_file","list_files","grep_search","write_file","fetch_paper_pdf","fetch_paper_metadata","openalex_search","crossref_search","semantic_scholar_search","arxiv_search","search_papers","multi_source_search","extract_pdf_text","extract_paper_sections","save_paper_note","finish_task"] |
| strict_tools | true |
| model_tier | standard |
| temperature | 0.15 |
| allowed_read_prefixes | ["user_inputs/paper-comparison/","literature/","drafts/","_runtime/skill_sessions/"] |
| allowed_write_prefixes | ["literature/"] |
| intake_tools | ["fetch_paper_pdf","fetch_paper_metadata","openalex_search","crossref_search","semantic_scholar_search","arxiv_search","search_papers","multi_source_search"] |
| outputs_expected | {"comparison_report":"literature/skill_paper_comparison.md","comparison_record":"literature/skill_paper_comparison.json"} |
| interaction | {"mode":"guided","language":"zh-CN","summary":"比较两篇或多篇已有论文阅读笔记、PDF、DOI/arXiv/OpenAlex 标识,或按明确主题和数量检索得到的论文;按问题、机制、方法、证据、局限和可组合性给出可追溯结论。","request_required":true,"request_prompt":"请说明要比较的研究问题、比较维度和希望得到的决策,例如选择基线、定位 gap 或合并机制。","example_request":"比较这些笔记卡如何处理 treatment heterogeneity,指出可作为 baseline 的方法和不能直接合并的假设。","required_inputs":[{"id":"sources","label":"待比较的来源清单","description":"一行一个来源;可为 workspace 相对的 paper note/PDF 路径、DOI、arXiv/OpenAlex ID、直接 PDF URL 或精确标题。也可在 request 中明确“主题 + 数量”,由系统先检索候选并保留检索记录;至少需要两项来源或一个明确主题数量请求。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":20,"example":"user_inputs/paper-comparison/sources.md"}],"optional_inputs":[{"id":"comparison_context","label":"比较框架或目标决策","description":"可选;给出固定比较轴、目标投稿方向、研究场景或希望避免的结论。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":10,"example":"user_inputs/paper-comparison/context.md"}],"outputs":[{"id":"comparison_report","label":"论文比较报告","path":"literature/skill_paper_comparison.md","description":"人类可读的比较矩阵、共同点/冲突、证据边界、可组合性和下一步建议。"},{"id":"comparison_record","label":"论文比较结构记录","path":"literature/skill_paper_comparison.json","description":"机器可读的来源路径、比较字段、相同/不同点、证据等级和未决项。"}]} |
Source-Grounded Paper Comparison
Read the source list first and classify each entry as a workspace path, PDF, DOI/arXiv/ OpenAlex identifier, URL, exact title, or an explicit topic-plus-count request. Resolve identifiers with metadata/search tools, and attempt PDF retrieval only to
user_inputs/paper-comparison/ when the user authorized that source. For a topic-plus- count request, write the exact query, returned candidates, selection rule, and access outcome before comparing; do not silently broaden the topic or replace unread papers with model memory. If a source is missing, malformed, inaccessible, or not a readable note/PDF, write a focused follow-up request and retain the limitation. Extract only what the actual sources support and retain each source path plus section/field anchors.
The report must distinguish observed facts, analytical comparisons, and proposed research implications. Compare at least: problem/unit of analysis, mechanism or design rationale, method/intervention, data/evaluation, evidence status, limitations, and nearest usable role (baseline, mechanism source, boundary case, or background). State whether a proposed merger is evidence-supported, merely plausible, or currently unsupported. Do not create BibTeX entries or say a source supports a claim without an anchor.