Skills for Real Engineers. Straight from my .agents directory.
science-literature-search
Agent skill fromyzlin286-wq/Deepscient
Use when starting a literature review, checking whether an idea is already published, or updating a topic search; covers query construction, multi-source coverage, deduplication, inclusion criteria, and what must be recorded so the search is reproducible.
How to install science-literature-search
npx skills add yzlin286-wq/Deepscient# or add only this skill to Claude Code: copy the folder
dsh-science/skills/science/science-literature-search/
# into
~/.claude/skills/science-literature-search/ (all projects)
.claude/skills/science-literature-search/ (one project)
1# 文献检索(可复现的检索策略)23## 何时用4- 课题开局要做现状梳理;5- 有 idea 需要做新颖性(scoop)检查;6- 距离上次检索超过 3 个月,需要增量更新。78## 前置检查91. `science_preflight` 已跑过,数据根与技能就位;102. 确认检索源:`mcp__paper__*`、`mcp__arxiv__*`、`mcp__zotero__*` 是否可用;不可用则退化为 `web_search` + `web_fetch` 打公开 API(Crossref / OpenAlex / Semantic Scholar / PubMed E-utilities),并在结果里标注"覆盖度下降"。113. 与用户确认三件事:**时间窗**、**语言**、**纳排标准**(哪些不算:预印本?综述?非同行评议?)。1213## 步骤141. **拆概念块**:把课题拆成 2–4 个概念块(PICO 或"方法 + 对象 + 指标"),每个块列同义词、缩写、上下位词。152. **构造检索式**:块内 OR,块间 AND。例:16 - `("reinforcement learning" OR RL) AND (protein OR "protein design") AND (diffusion OR "generative model")`17 - 中文库另写一份(CNKI/万方用主题词 + 同义词)。183. **多源执行**:至少 2 个源;确保 arXiv 覆盖预印本、PubMed 覆盖生物医学、OpenAlex 覆盖跨库引用。每个源记录:库名、检索式、时间、结果数。194. **去重**:按 DOI → arXiv id → 标题指纹(去标点、小写、去副标题)+ 首作者 + 年份 三路去重。落库走 `science_lit_search`,重复项不要重复入库。205. **两阶段筛选**:先标题+摘要(记录排除理由代码),再全文。排除理由必须可统计(PRISMA 流程图用)。216. **滚雪球**:对纳入的每篇做前向(被引)+ 后向(参考文献)各一轮,重点找 2 篇最相关的。227. **落库与汇报**:`science_lit_search` 落 `paper` 表;汇报只写:源与检索式、命中数、纳入数、最相关的 5 篇 + 一句话差异。2324## 产出与落库25- `paper` 表:doi / arxiv_id / title / authors / year / venue / url / pdf_path / source / added_at26- 检索过程写进项目 `logs/search-<date>.md`(检索式、库、时间、命中数),供复现27- `science_novelty`:最接近的 10 篇 + 与 idea 的差异点2829## 常见失败与降级30- **MCP 全挂**:走公开 API 兜底,明确告诉用户"只覆盖 X 库"。31- **命中 >500**:先收紧时间窗或加限定词,不要一次性拉全量(token 与速率都会崩)。32- **完全无命中**:这本身是结论(可能是真空白,也可能是检索式写得过窄);换 2 套检索式再确认,然后如实报告"未检索到直接相关工作"。33- **警惕**:不要凭记忆补引用。任何作者/年份/DOI 都必须来自库内记录。
Excerpt shown under the repository's MIT license.
More skills in Deepscient
- ▸science-agent-scheduling
- ▸science-campaign-design
- ▸science-consensus-report
- ▸science-discipline-methods
- ▸science-experiment-runbook
- ▸science-figure-for-journal
- ▸science-frontier-to-paper
- ▸science-fulltext-hunting
- ▸science-journal-intel
- ▸science-journal-selection
- ▸science-manuscript-formatting
- ▸science-org-roles
Similar skill repositories
An agentic skills framework & software development methodology that works.
A single CLAUDE.md file to improve Claude Code behavior, derived from Andrej Karpathy's observations on LLM coding pitfalls.
Public repository for Agent Skills
Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.
Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
