Scientific Agent Skills是一套专门为科学研究、工程分析、金融建模、技术写作设计的Agent技能组合。它不是单一工具，而是将科研工作流的每个环节都封装成可组合的技能单元。
凌晨4点17分，你的论文截止日期就在明天。这时你需要的是一个能理解学术语境、能检索文献、能分析数据、能生成规范引用的Agent——这就是Scientific Agent Skills要解决的问题。
arXiv、PubMed、Google Scholar自动检索与摘要生成
统计检验、回归分析、可视化图表生成
假设生成、变量控制、样本量计算
结构化写作、引用管理、格式规范
自动检测方法论问题、逻辑漏洞
高质量图表、交互式数据展示
// OpenClaw文献检索技能调用
{
"skill": "literature_search",
"query": "transformer architecture attention mechanism",
"sources": ["arxiv", "pubmed", "semantic_scholar"],
"filters": {
"year_range": [2020, 2026],
"min_citations": 50,
"open_access": true
},
"output": {
"format": "structured",
"include_abstract": true,
"include_citations": true,
"max_results": 20
}
}# OpenClaw数据分析工作流
from openclaw.skills import DataAnalysis
# 初始化分析器
analyzer = DataAnalysis(
data_source="experiments.csv",
output_format="markdown_report"
)
# 执行分析管道
report = analyzer.run_pipeline([
"descriptive_stats", # 描述性统计
"normality_test", # 正态性检验
"correlation_analysis", # 相关性分析
"regression", # 回归分析
"visualization" # 可视化
])
# 生成论文级图表
analyzer.export_figures(
format="svg",
dpi=300,
style="academic"
)// 实验设计技能配置
{
"skill": "experiment_designer",
"research_question": "新药X对阿尔茨海默病患者认知功能的影响",
"variables": {
"independent": ["药物剂量", "给药频率"],
"dependent": ["MMSE评分", "ADAS-Cog评分"],
"controlled": ["年龄", "病程", "基线认知水平"]
},
"design": {
"type": "randomized_controlled_trial",
"groups": 4,
"sample_size_calculation": {
"effect_size": 0.5,
"power": 0.8,
"alpha": 0.05
}
},
"output": {
"protocol_template": "CONSORT",
"ethics_checklist": true
}
}# OpenClaw学术写作技能
## 结构化写作流程
1. **摘要生成** - 基于研究结果自动生成结构化摘要
2. **引言撰写** - 文献综述+研究空白+研究问题
3. **方法描述** - 可复现性导向的方法论撰写
4. **结果呈现** - 图表+统计结果+关键发现
5. **讨论生成** - 结果解释+理论贡献+局限性
6. **引用管理** - 自动生成规范引用列表
## 支持的引用格式
- APA 7th Edition
- MLA 9th Edition
- Chicago Style
- IEEE
- Nature/Science格式
- 中文GB/T 7714# SKILL.md - 科研助手完整配置
name: scientific-research-assistant
description: 端到端科研助手技能包
skills:
# 文献检索
- name: literature_search
sources: [arxiv, pubmed, semantic_scholar, google_scholar]
features:
- citation_network_analysis
- related_works_discovery
- author_profiling
# 数据分析
- name: data_analysis
capabilities:
- statistical_tests
- machine_learning
- time_series_analysis
- survival_analysis
# 可视化
- name: visualization
output_formats: [svg, png, pdf, interactive_html]
styles: [academic, nature, science, ieee]
# 写作辅助
- name: academic_writing
templates: [research_paper, review_article, grant_proposal]
languages: [en, zh]
# 翻译与润色
- name: language_polish
services: [grammar_check, style_improve, translation]
workflows:
- name: full_research_pipeline
steps:
- literature_search: { query: "${research_topic}" }
- literature_review: { papers: "$[0].results" }
- experiment_design: { hypothesis: "${hypothesis}" }
- data_analysis: { data: "${experiment_data}" }
- visualization: { results: "$[3]" }
- paper_drafting: { sections: [intro, methods, results, discussion] }
- language_polish: { draft: "$[5]" }