| name | literature-resource-scout |
| description | Identify concrete datasets, benchmarks, baselines, code artifacts, and reproducibility constraints from a research brief and existing literature artifacts. Use when a researcher needs a resource feasibility report without treating unverified web snippets as evidence. |
| tools | ["read_file","write_file","list_files","multi_source_search","semantic_scholar_search","arxiv_search","openalex_search","crossref_search","finish_task"] |
| strict_tools | true |
| model_tier | standard |
| temperature | 0.3 |
| allowed_read_prefixes | ["user_inputs/literature-resource-scout/","literature/","ideation/","_runtime/skill_sessions/"] |
| allowed_write_prefixes | ["literature/"] |
| outputs_expected | {"report":"literature/skill_resource_scout.md","inventory":"literature/skill_resource_inventory.json"} |
| interaction | {"mode":"guided","language":"zh-CN","summary":"为一个研究问题盘点真实可访问的基准、数据、基线、代码和复现风险；所有条目必须带来源和访问状态。","request_required":true,"request_prompt":"请说明方法或任务、希望寻找的数据/基线、可用算力与许可证限制。","example_request":"为当前研究问题寻找公开数据集、可复现 benchmark 和强基线，优先 Apache/MIT 许可证。","required_inputs":[{"id":"research_brief","label":"资源搜索问题","description":"说明任务、研究对象、必需资源类型和排除条件；可以是独立 brief 或已有文献缺口说明。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":60,"example":"user_inputs/literature-resource-scout/brief.md"}],"optional_inputs":[{"id":"existing_synthesis","label":"已有文献综合","description":"可选；用于避免重复检索，并把资源与文献卡的 section 对齐。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":80,"example":"literature/synthesis.md"}],"outputs":[{"id":"report","label":"资源可行性报告","path":"literature/skill_resource_scout.md","description":"解释哪些资源可用、证据来源、限制、风险和建议优先级。"},{"id":"inventory","label":"结构化资源清单","path":"literature/skill_resource_inventory.json","description":"每项的 URL/论文标识、许可证线索、访问状态、用途和未核验字段。"}]} |
Literature Resource Scout
Search only after reading the verified brief and optional synthesis. Use source-returning tools; distinguish a paper's claim from an accessible dataset or repository. Never infer a license, benchmark availability, metric, or code release when the source response does not show it.
For every retained resource, record a stable identifier or URL, source type, linked paper, intended research role, access status, licence evidence or unverified, compatibility constraints, and a reason to keep, defer, or reject it. Keep uncertain candidates as leads, not evidence. Write the complete Markdown report and JSON inventory; end with a prioritized, actionable shortlist.
