research-literature-scout
Use when user requests paper literature retrieval evidence mapping query planning and source verification. Works within the general Research Idea Discovery plugin and any user-selected field.
- License 许可
- MIT
- 最近更新
- 2026-10-13
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- research-literature-scout
- description
- Use when user requests paper literature retrieval evidence mapping query planning and source verification. Works within the general Research Idea Discovery plugin and any user-selected field.
03 SKILL.md 正文
文献调查 + Idea 后主动检索数据集
先阅读可获得论文资料,构建带来源的失败现象/机制差异 Evidence Map;不必先找数据才能让 Hypothesis Lab 产生 IDEA_SEED。当收到 Seed 后,根据其所需模态、字段、标签、评测/最小证伪实验,主动搜索现有数据/父数据、官方 benchmark、论文项目页、许可、代码、可复现 baseline。优先近邻论文的数据集,扩展跨平台/官方仓库;条件允许时通过 ChatGPT web 工具或宿主中的 scripts/dataset_discovery.py 检索线索,再逐项核查。
每个 Seed 输出 dataset-search-ledger.json:搜索词、日期、至少若干潜在数据来源、官方链接与元数据/下载/访问证据、字段标签、许可、划分、基线、与 E0 适配、低成本衍生方案。若找不到,NO_DATA_PATH 并回到种子生成;不得要求用户预先给已核实数据集。
区分 METADATA_ONLY / PAGE_CHECKED / ACCESS_CONFIRMED / DERIVATION_FEASIBLE / BLOCKED / UNKNOWN。元数据不等于下载成功,离线检查不等于真实许可已核验;没有证据就不能通过 D0。不可声称绝对不存在数据集。
当常规文献搜索达到实质性证据瓶颈或用户请求深度调研时,交给 research-deep-research-bridge 构建有具体缺口的可复制 Prompt,记录已检查文献/查询与不确定性。外部报告回传后作为二手来源,不直接升级 E2/E3 全文证据等级。
04 同仓库更多 Skills
research-review-board
暂无简介
research-proposal-generator
暂无简介
research-problem-validation
Validate research problems before generating ideas.
research-problem-scout
Use at the beginning of scientific idea discovery to find real, important, falsifiable research failures from actual sources; reject weak motivations before architectures and novelty-hunting. Includes low-cost problem-evidence scouting and a hard Motivation-First Gate.
research-pilot-designer
Use when user requests falsification pilot experiment design hypothesis testing research planning. Works within the general Research Idea Discovery plugin and any user-selected field.
research-paper-anchor-scout
Critically read a small adversarial group of anchor papers to discover real, falsifiable research problems from failures, successful-method mechanisms, contradictions and deployment assumptions; reuse verified assets without blindly modifying the previous network.