| name | paper-section-evidence |
| description | Answer a bounded research question from exact sections of one uploaded paper PDF, preserving page and section anchors and evidence boundaries. Use when a researcher needs a precise method, result, limitation, dataset, or claim check without a full literature note card. |
| tools | ["read_file","write_file","process_seed_paper","extract_paper_sections","extract_pdf_text","finish_task"] |
| strict_tools | true |
| model_tier | standard |
| temperature | 0.1 |
| allowed_read_prefixes | ["user_inputs/paper-section-evidence/","user_seeds/","literature/","_runtime/skill_sessions/"] |
| allowed_write_prefixes | ["user_seeds/","literature/"] |
| outputs_expected | {"evidence_report":"literature/skill_section_evidence.md","evidence_record":"literature/skill_section_evidence.json"} |
| interaction | {"mode":"guided","language":"zh-CN","summary":"上传一篇 PDF 并提出具体问题，从论文中的相关内容提取并标注位置的证据、可采用的谨慎表述和不确定项，而非泛泛总结整篇论文。","request_required":true,"request_prompt":"请明确要核验的问题，例如方法机制、数据集、实验设置、报告数字、局限，或某个待写 claim 是否得到支持。","example_request":"从该论文的 Method、Experiments 和 Limitations 核验其异质 treatment 机制、实验数据和可用于 Related Work 的保守表述。","required_inputs":[{"id":"paper_pdf","label":"单篇研究论文 PDF","description":"上传真实可读 PDF；Skill 只对实际读取到的论文内容作出结论。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":1024,"example":"user_inputs/paper-section-evidence/paper.pdf"}],"optional_inputs":[{"id":"known_identifier","label":"已知 DOI、arXiv ID 或题名","description":"可选；用于标注来源，不会替代对 PDF 内容的实际读取。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":10,"example":"user_inputs/paper-section-evidence/context.md"}],"outputs":[{"id":"evidence_report","label":"论文内容核验报告","path":"literature/skill_section_evidence.md","description":"对每个问题列出查阅的论文内容、直接证据、可采用表述、不能延伸的结论和需要补读的位置。"},{"id":"evidence_record","label":"论文内容核验结构记录","path":"literature/skill_section_evidence.json","description":"机器可读的问题、读取范围、论文位置与页码、证据等级、结论和不确定性。"}]} |
Paper Section Evidence
Read the request and register the uploaded PDF through process_seed_paper with
source="pdf_path" and the declared input path. Translate the request into a small set of target sections, then call extract_paper_sections before reading broader text.
Use extract_pdf_text only to resolve a concrete missing fact, and record its page-coverage or truncation status.
Answer only from text actually extracted from the PDF. For every substantive answer, retain the section heading and page or locator returned by the tool. Separate direct source facts, cautious interpretation, and unsupported requests. A method description does not prove a reported result; an abstract-only extraction does not justify a mechanism or causal claim. Do not manufacture quotations, numeric values, a DOI, or a BibTeX entry.
Write both declared outputs even when the requested evidence is unavailable. In that case name the exact unreadable or missing section and formulate the smallest useful human follow-up. This Skill is intentionally narrower than pdf-note-card: it must not claim to create a complete T3 note card.
