Evidence-first biomedical manuscript review, from the claimed conclusion back to the evidence.
中文说明见下半部分；English documentation follows the Chinese documentation.
BioEvidenceChain Lite 是一套面向生物医学原创研究论文的结论优先证据审查框架，也是一项可安装的 Agent Skill。
它不从方法学清单开始，而是先还原作者真正想证明的结论，再围绕这条主线判断创新性、科学意义、最低证据、现有证据与关键缺口，最后回到结论，说明：
- 已经证明了什么；
- 还没有证明什么；
- 若不补实验，结论应收缩到哪里。
当前版本：v2.1.0
常见的 AI 审稿容易把创新性、方法、统计、Figure 和 Discussion 拆成互不相干的检查表，最后得到一串看似全面、却没有主次的问题。
BioEvidenceChain 使用同一张结论—证据地图贯穿审查：
核心结论
↓
创新性与科学意义
↓
最低证据与逐图核查
↓
数据、方法与论证风险
↓
Discussion、局限与文献定位
↓
重新划定结论边界
核心原则是：每一个重要问题都必须绑定到具体结论，每一项补充建议都必须说明它要排除哪个替代解释。
- 还原论文的结论主线及关键节点；
- 按主张检索既有研究，判断真正的新颖性与科学意义；
- 先列最低证据，再按 Figure 或结果模块核查证据；
- 识别因果越级、替代解释、模型边界和人类证据等级；
- 检查文稿层面的实验单位、伪重复、多重比较、效应量与一致性风险；
- 审查 Introduction 是否建立了真实知识缺口，Discussion 是否处理矛盾证据与局限；
- 根据使用者指令语言，生成内容一致的中文或英文 Markdown 报告与 A4 PDF。
报告语言由使用者的指令语言决定，而不是由论文语言决定：
- 中文提示词 → 输出完整中文报告；
- 英文提示词 → 输出完整英文报告；
- 明确指定输出语言时，以指定语言为准；
- 只有使用者明确要求时，才输出双语报告。
标题、表头、证据状态、解释文字和 PDF 排版均使用同一种选定语言。
| BioEvidenceChain Lite 不提供 | 原因或边界 |
|---|---|
| 正式 Reviewer Comments 或编辑密信 | Lite 定位为科学性初审与投稿前自查 |
| Accept / Minor / Major / Reject 建议 | 不替代期刊编辑裁定 |
| 目标期刊匹配与发表概率判断 | 期刊适配属于其他工作流 |
| 原始图像、FCS、个体数据或代码审计 | Lite 仅评价论文及其补充材料中可见、可报告的证据 |
| 图像取证或学术不端认定 | 只描述可观察的不一致及进一步核查所需材料 |
| 综述、修回稿复审或 rebuttal | 本 Skill 只面向生物医学原创研究初审 |
至少提供完整论文 PDF、主图、图注和表格；如有 Supplementary Materials，建议一并上传。
安装 Skill 后输入：
请使用 BioEvidenceChain Lite 审查附件中的生物医学原创研究论文。
重点问题：［可不填］
请先准确提炼作者意图表达的结论主线，不要在这一步评价证据是否充分；随后评价创新性和科学意义，并按 Figure 核查最低证据、现有证据和关键缺口。最终同时输出内容一致的中文 Markdown 报告和 A4 PDF，PDF 四边页边距 2.54 cm；不生成正式审稿意见或接收、修改、拒稿建议。若 PDF 生成失败，仍须交付 Markdown 并说明失败原因。
- 在本仓库右侧 Releases 下载
BioEvidenceChain-Lite-v2.1.0.zip。 - 打开侧边栏中的 Skills；若当前版本提供导入入口，上传完整 ZIP。
- 也可以调用
@skill-creator，要求其根据该 ZIP 创建或导入 Skill。 - 安装后确认名称显示为
BioEvidenceChain Lite。
Standalone Skills 当前主要面向 ChatGPT 桌面版、Codex CLI 和 IDE 扩展；网页端与移动端的直接安装能力取决于产品和工作区是否开放相应入口。参见 OpenAI 官方 Skills 文档。
下载并解压 Release ZIP，把完整的 bioevidencechain-lite 文件夹放入：
- 个人范围：
$HOME/.agents/skills/ - 当前项目：
项目目录/.agents/skills/
然后在 Codex 中输入：
使用 $bioevidencechain-lite 审查附件中的论文。
Codex 通常会自动检测 Skill 变更；如未出现，可重新启动。参见 OpenAI 官方安装说明。
下载并解压 Release ZIP，把完整文件夹放入：
- 个人范围：
$HOME/.claude/skills/ - 当前项目：
项目目录/.claude/skills/
随后输入 /bioevidencechain-lite，或直接要求 Claude 使用该 Skill。参见 Claude Code Skills 文档。
可以直接从本仓库安装指定版本：
openclaw skills install git:Comike-Li/BioEvidenceChain@v2.1.0 --as bioevidencechain-lite也可以下载 Release ZIP、解压后从本地安装：
openclaw skills install ./bioevidencechain-lite --as bioevidencechain-lite增加 --global 可供本机所有 OpenClaw 智能体使用。参见 OpenClaw Skills 文档。
若平台支持 Agent Skills、SKILL.md 或自定义 Skill 导入，上传完整 ZIP 或文件夹即可。若没有专门入口，可将：
SKILL.md设为主系统指令或工作流入口；references/中的规则文件设为关联知识文件；- 要求智能体先读取
SKILL.md，再按其中指引加载相应参考文件。
不要只复制 SKILL.md；必须同时保留 references/、agents/、版本与许可文件。
- 报告摘要
- 结论主线
- 创新性与科学意义
- 证据链评价
- 论文数据与方法学初评
- Introduction 与 Discussion 评价
- 这篇论文最需要解决的问题
- 建议补充阅读的文献
- 审查范围与限制
BioEvidenceChain/
├── SKILL.md
├── VERSION
├── LICENSE.md
├── CHANGELOG.md
├── CITATION.cff
├── README.md
├── agents/
│ └── openai.yaml
└── references/
├── conclusion-loop.md
├── novelty-significance.md
├── evidence-reliability.md
└── output-contract.md
SKILL.md 是入口；其余规则按任务阶段渐进加载，避免一次性塞入全部上下文。
BioEvidenceChain Lite 用于辅助研究者拆解论证、检索文献和发现证据缺口，不能代替作者、审稿人或编辑承担最终学术判断。文献检索结果会受数据库和全文访问范围影响，涉及重要结论时应由研究者复核原始来源。
如在研究、教学或公开项目中使用本框架，可点击仓库右侧的 Cite this repository 获取由 CITATION.cff 生成的引用信息。
如发现漏检关键证据、结论提炼偏差、过度要求实验、跨模型外推或安装问题，请通过 GitHub Issues 反馈。为保护隐私，请勿上传未公开稿件、原始数据、审稿邀请或其他受保密义务约束的材料。
BioEvidenceChain Lite 以 CC BY-NC 4.0 公开发布：允许在署名、注明修改且保持非商业用途的前提下分享和改编；商业使用、付费服务或商业产品集成需另行获得许可。
CC BY-NC 4.0 是非商业内容许可，并非 OSI 认可的软件开源许可证。因此对外传播建议使用“公开发布”“源码公开”或“非商业许可”，不要表述为“完全开源”。
BioEvidenceChain Lite is a conclusion-first evidence-auditing framework and installable Agent Skill for original biomedical research manuscripts.
Instead of starting from a generic methods checklist, it first reconstructs what the authors intend to prove. It then evaluates novelty, scientific significance, minimum required evidence, evidence actually supplied, and decisive gaps. Finally, it returns to the main conclusion and states:
- what the manuscript has established;
- what it has not established;
- how the conclusion should be narrowed if no additional experiment is added.
Current version: v2.1.0
AI-generated manuscript reviews often treat novelty, methods, statistics, figures, and Discussion as separate checklists. The result may look comprehensive while failing to distinguish conclusion-breaking gaps from local defects.
BioEvidenceChain uses one conclusion–evidence map throughout the review:
Claimed conclusion
↓
Novelty and scientific significance
↓
Minimum evidence and figure-level audit
↓
Data, methods, and reasoning risks
↓
Discussion, limitations, and literature position
↓
Recalibrated conclusion boundary
Every material issue must be linked to a specific conclusion, and every proposed experiment must state which credible alternative it is intended to distinguish.
- Reconstructs the main conclusion and its claim nodes or links.
- Searches prior work claim by claim to assess defensible novelty and significance.
- Lists minimum evidence before auditing each figure or coherent result module.
- Separates observation, association, causation, mechanism, disease relevance, human relevance, and translation.
- Identifies manuscript-level risks involving experimental units, pseudoreplication, multiplicity, effect size, and internal consistency.
- Assesses whether the Introduction establishes a real gap and whether the Discussion addresses contrary evidence, alternatives, and limitations.
- Produces matching Markdown and A4 PDF reports in Chinese or English.
The report follows the language of the user's instruction, not the language of the manuscript:
- Chinese request → Chinese report.
- English request → English report.
- An explicit language instruction takes priority.
- A bilingual report is produced only when explicitly requested.
All headings, tables, status labels, explanations, and PDF typography use the selected language consistently.
| BioEvidenceChain Lite does not provide | Boundary |
|---|---|
| Formal reviewer comments or confidential editor letters | Lite is for scientific screening and pre-submission self-review |
| Accept / Minor / Major / Reject recommendations | It does not replace editorial decisions |
| Journal matching or acceptance-probability estimates | Journal-specific assessment belongs to another workflow |
| Raw-image, FCS, individual-level data, or code audits | Lite assesses evidence visible or reported in the manuscript and supplements |
| Image forensics or misconduct findings | It reports observable inconsistencies and the source material needed for verification |
| Review articles, revised-manuscript re-review, or rebuttals | This Skill targets initial assessment of original biomedical research |
Provide the complete manuscript PDF, main figures, legends, and tables. Include supplementary materials when available.
Then use:
Please use BioEvidenceChain Lite to assess the attached original biomedical research manuscript.
Priority question: [optional]
First reconstruct the conclusion mainline exactly as the authors intend it, without judging evidence sufficiency at that stage. Then assess novelty and scientific significance, list the minimum evidence required for the central claims, and audit the manuscript figure by figure. Deliver matching English Markdown and A4 PDF reports with 2.54 cm margins. Do not produce formal reviewer comments or an Accept, Minor, Major, or Reject recommendation. If PDF generation fails, still deliver the complete Markdown report and state the reason.
- Download
BioEvidenceChain-Lite-v2.1.0.zipfrom Releases. - Open Skills in the sidebar and, if your version provides an import control, upload the complete ZIP.
- Alternatively, invoke
@skill-creatorand ask it to create or import the Skill from the ZIP. - Confirm that the installed name is
BioEvidenceChain Lite.
Standalone Skills are currently intended primarily for the ChatGPT desktop app, Codex CLI, and IDE extension. Direct installation on web or mobile depends on product and workspace availability. See the official OpenAI Skills documentation.
Download and extract the Release ZIP. Place the complete bioevidencechain-lite folder in either:
- Personal scope:
$HOME/.agents/skills/ - Current project:
<project>/.agents/skills/
Invoke it with:
Use $bioevidencechain-lite to assess the attached manuscript.
Codex normally detects Skill changes automatically; restart if it does not appear. See the official OpenAI installation documentation.
Download and extract the Release ZIP. Place the complete folder in either:
- Personal scope:
$HOME/.claude/skills/ - Current project:
<project>/.claude/skills/
Then enter /bioevidencechain-lite, or ask Claude to use the Skill. See the Claude Code Skills documentation.
Install a tagged version directly from this repository:
openclaw skills install git:Comike-Li/BioEvidenceChain@v2.1.0 --as bioevidencechain-liteOr download and extract the Release ZIP, then install locally:
openclaw skills install ./bioevidencechain-lite --as bioevidencechain-liteAdd --global to make it available to all local OpenClaw agents. See the OpenClaw Skills documentation.
If the platform supports Agent Skills, SKILL.md, or custom Skill import, upload the complete ZIP or folder. Otherwise:
- Use
SKILL.mdas the main system instruction or workflow entry point. - Attach the files under
references/as linked knowledge files. - Require the agent to read
SKILL.mdfirst and load the referenced files in the instructed order.
Do not copy SKILL.md alone. Preserve the references/, agents/, version, and license files.
- Executive Summary
- Conclusion Mainline
- Novelty and Scientific Significance
- Evidence-Chain Assessment
- Preliminary Data and Methods Assessment
- Introduction and Discussion Assessment
- Highest-Priority Issues
- Recommended Key Literature
- Scope and Limitations
BioEvidenceChain Lite helps researchers examine reasoning, literature, and evidence gaps. It does not replace the final scientific judgment of authors, reviewers, or editors. Literature findings depend on database coverage and full-text access; consequential claims should be checked against the primary sources.
Use Cite this repository on the right side of the GitHub page to obtain citation metadata generated from CITATION.cff.
Please use GitHub Issues to report missed evidence, inaccurate reconstruction of the main conclusion, excessive experimental requests, unsupported cross-model extrapolation, or installation problems. Do not upload confidential manuscripts, raw data, peer-review invitations, or other materials subject to confidentiality obligations.
BioEvidenceChain Lite is released under CC BY-NC 4.0. Sharing and adaptation are permitted with attribution, indication of changes, and noncommercial use. Commercial use, paid services, or integration into commercial products requires separate permission.
CC BY-NC 4.0 is a noncommercial content license and is not an OSI-approved open-source software license. Describe the project as “publicly released,” “source-available,” or “noncommercially licensed,” rather than “fully open source.”