| name | literature-evidence-matrix |
| description | Build a compact, source-traceable comparison matrix from a bounded set of ResearchOS paper notes, PDF note cards, or identifier records. Use when a researcher needs a review-ready evidence table before synthesis, survey writing, or idea selection. |
| tools | ["read_file","list_files","grep_search","write_file","finish_task"] |
| strict_tools | true |
| model_tier | standard |
| temperature | 0.1 |
| allowed_read_prefixes | ["user_inputs/literature-evidence-matrix/","literature/","_runtime/skill_sessions/"] |
| allowed_write_prefixes | ["literature/"] |
| outputs_expected | {"matrix_csv":"literature/skill_evidence_matrix.csv","matrix_report":"literature/skill_evidence_matrix.md","matrix_record":"literature/skill_evidence_matrix.json"} |
| interaction | {"mode":"guided","language":"zh-CN","summary":"将一组已有论文阅读笔记或记录整理为带可回查来源位置、证据等级和空白项的比较矩阵，供综述与 idea 使用。","request_required":true,"request_prompt":"请说明矩阵要服务的主题、比较字段，以及是否优先方法 taxonomy、实验设计、机制或局限。","example_request":"为当前研究主题的综述生成方法/数据/任务/指标/机制/局限矩阵，并标注哪些行只允许作为背景。","required_inputs":[{"id":"source_list","label":"文献来源路径清单","description":"一行一个 workspace 相对路径，指向结构化 paper note、PDF note card 或 identifier record；建议 3-30 项。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":20,"example":"user_inputs/literature-evidence-matrix/sources.md"}],"optional_inputs":[{"id":"matrix_schema","label":"自定义比较字段","description":"可选；指定列、术语、目标读者和哪些字段必须保留为空而不是猜测。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":10,"example":"user_inputs/literature-evidence-matrix/schema.md"}],"outputs":[{"id":"matrix_csv","label":"文献证据矩阵 CSV","path":"literature/skill_evidence_matrix.csv","description":"一行一篇来源，包含可追溯字段、证据等级和来源路径；未知值保留为空或明确 unknown。"},{"id":"matrix_report","label":"矩阵覆盖报告","path":"literature/skill_evidence_matrix.md","description":"说明输入覆盖、字段缺失、弱证据比例、可用于综述的边界和建议补检方向。"},{"id":"matrix_record","label":"矩阵结构记录","path":"literature/skill_evidence_matrix.json","description":"机器可读的矩阵 schema、行来源、空白字段、证据状态和生成限制。"}]} |
Literature Evidence Matrix
Validate each requested source path before extracting fields. Use source-native labels where possible and record the path and note section/field that supplied every nontrivial cell. A missing field must remain blank or unknown; never infer a dataset, metric, causal mechanism, result, or citation from a title alone.
Write a CSV that is valid RFC-style CSV: quote cells containing commas, quotes, or new lines. Keep the table compact enough to scan, while the Markdown report carries longer explanations, weak-evidence warnings, coverage counts, and proposed section-specific follow-up searches. Clearly distinguish full-text/partial-text/abstract-only/metadata records and prevent the latter two from becoming strong evidence rows.
