literature-comparison-studio

name	literature-comparison-studio
description	Turn a researcher’s comparison question plus DOI, title, PDF, paper-note, or source list into a traceable pairwise, corpus, or method-family comparison with evidence anchors and a decision-ready report. Use when paper-comparison alone is too manual because source preparation is incomplete.
tools	["read_file","write_file","write_structured_file","list_files","grep_search","fetch_paper_metadata","openalex_search","crossref_search","semantic_scholar_search","arxiv_search","search_papers","multi_source_search","fetch_paper_pdf","extract_pdf_text","extract_paper_sections","save_paper_note","ask_human","update_skill_workflow","finish_task"]
strict_tools	true
model_tier	standard
temperature	0.15
allowed_read_prefixes	["user_inputs/literature-comparison-studio/","literature/","user_seeds/","_runtime/skill_sessions/"]
allowed_write_prefixes	["user_inputs/literature-comparison-studio/","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/comparison_studio.md","comparison_csv":"literature/comparison_studio.csv","comparison_record":"literature/comparison_studio.json","comparison_claims":"literature/comparison_claims.md","workflow_manifest":"literature/comparison_studio_manifest.json"}
interaction	{"mode":"guided","language":"zh-CN","summary":"输入比较目的与 DOI、标题、PDF 或论文阅读笔记，或只给明确主题和目标篇数；系统会检索、解析和尽量补齐可读来源，标注可回查的证据位置，并输出可用于综述、baseline 选择或机制定位的对比结论。","request_required":true,"request_prompt":"请说明要比较什么、为何比较、希望得到什么决策，以及采用两篇、指定语料还是方法家族模式。","example_request":"检索并比较 treatment heterogeneity 主题的 4 篇近年论文，优先获取可读全文，判断哪些可作为 baseline，哪些机制不能直接合并。","required_inputs":[{"id":"comparison_request","label":"比较问题与决策目标","description":"说明比较对象或其标识、比较目的、需要的轴和禁止推断的内容；可给 DOI/标题列表，也可明确主题、目标篇数、时间范围和来源偏好。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":50,"example":"user_inputs/literature-comparison-studio/request.md"}],"optional_inputs":[{"id":"source_manifest","label":"来源文件或标识清单","description":"可选；一行一个 DOI、arXiv/OpenAlex ID、标题、PDF URL、workspace PDF、paper note 或 record 路径。缺失项会进入定向检索/解析或精确补料，而不是被模型记忆替代。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":10,"example":"user_inputs/literature-comparison-studio/sources.md"}],"outputs":[{"id":"comparison_report","label":"文献比较报告","path":"literature/comparison_studio.md","description":"人类可读的比较矩阵、共同点、冲突、局限、可组合性和决策建议。"},{"id":"comparison_csv","label":"文献比较矩阵 CSV","path":"literature/comparison_studio.csv","description":"每行一个来源、每列一个显式比较轴，未知项保持 unknown。"},{"id":"comparison_record","label":"文献比较结构记录","path":"literature/comparison_studio.json","description":"来源标识、paper-card 路径、section 锚点、证据等级和结论限制。"},{"id":"comparison_claims","label":"可用比较主张","path":"literature/comparison_claims.md","description":"允许写入综述/Related Work 的措辞、来源锚点与不可主张边界。"},{"id":"workflow_manifest","label":"比较工作流清单","path":"literature/comparison_studio_manifest.json","description":"输入解析、下载/读取状态、未解析来源、阶段产物和恢复信息。"}]}
workflow	{"kind":"integrated","summary":"从比较问题和不完整来源开始，先解析并补齐可读证据，再给出来源锚定的比较与决策边界。","phases":[{"id":"comparison_contract","label":"比较合同与对象解析","objective":"确认比较模式、决策问题、对象和必须保留的未知项。","operations":"[Truncated]"},{"id":"source_readiness","label":"来源解析与补料","objective":"尝试解析 DOI/标题/PDF/笔记卡，或按已授权主题和篇数检索候选；无法解析或无全文时保留访问状态并向用户发出精确补料请求。","operations":"[Truncated]","human_gate":true},{"id":"evidence_extraction","label":"证据卡与 section 锚点","objective":"为关键比较轴建立 paper-card 或 section 级锚点，严格标注证据等级。","operations":"[Truncated]"},{"id":"comparison_and_audit","label":"对比、限制与决策","objective":"输出共同点、差异、冲突、可组合性和可用主张，不用标题/摘要推断实验细节。","operations":"[Truncated]"}]}

Literature Comparison Studio

Use the source manifest or an explicitly scoped topic-plus-count request only as a starting point. For topic retrieval, persist the query, requested count, candidate list, selection rule, and access results before extracting evidence. A DOI/title result is metadata evidence; a PDF is not evidence until the relevant text/section has been read. Create or reuse lightweight paper cards only within the declared workspace paths. If a decision-critical comparison axis cannot be sourced, leave it unknown, place it in the missing-evidence ledger, and ask the human for an upload or a narrower question.

Every nontrivial matrix cell and every statement in comparison_claims.md needs a source file plus note field or section anchor. Explicitly distinguish observed facts, analytical comparison, and proposed research implication. Never create a dataset, metric, baseline, result, or citation from a title, common convention, or model memory.