Skills for Real Engineers. Straight from my .agents directory.
science-consensus-report
Agent skill fromyzlin286-wq/Deepscient
Use when several agents or votes produced different answers and the result must be merged into one defensible statement — when running the three-vote rule, when weighting preprints against peer-reviewed evidence, when escalating an unresolved disagreement, or when writing the evidence table and gap list that a campaign report must carry.
How to install science-consensus-report
npx skills add yzlin286-wq/Deepscient# or add only this skill to Claude Code: copy the folder
dsh-science/skills/science/science-consensus-report/
# into
~/.claude/skills/science-consensus-report/ (all projects)
.claude/skills/science-consensus-report/ (one project)
1# 共识与报告:把多份结果合成一个可辩护的结论23## 三票规则45- 默认 `votes = 3`,`voteThreshold = 0.66`:三票中至少两票一致才采用。6- 三票必须**相互独立**:不同分片、不同输入顺序、不得共享中间摘要。7 同一上下文里连问三次不算三票。8- **只对结论型判断投票**:纳入/排除、新颖性、审稿意见分级、字段抽取。9 统计量、图表、格式转换属于确定性产物,一票即可,多票是浪费。10- 三票全不同 → 不合并、不取平均,直接按"分歧"上报。11- 两票一致但第三票给出反例证据 → 不按多数,转人工/上级裁决。1213## 证据权重1415投票之前先给证据定级,权重高的票不该被数量压过。1617| 层级 | 证据 | 权重 | 说明 |18| --- | --- | --- | --- |19| E1 | 本课题的原始数据 + 可重跑运行记录 | 1.0 | 有 `science_exp_run` 溯源才成立 |20| E2 | 同行评议论文的正文/表格原文 | 0.8 | 需 DOI + 页码/段落定位 |21| E3 | 预印本、会议摘要、注册平台记录 | 0.5 | 必须标注"未经同行评议" |22| E4 | 数据库条目、期刊官网元数据 | 0.4 | 适合事实字段,不适合结论 |23| E5 | 二手综述、博客、厂商文档 | 0.2 | 只能作为线索,不作断言依据 |2425规则:2627- **E3 不得单独支撑结论**。只有预印本支持时,写成"目前仅见预印本报告"。28- **E5 只出现在"待核实"栏**,不进正文结论。29- 同一结论有 E1 与 E2 时,优先 E1,并说明与文献是否一致。30- 权重用于解释分歧,不用于算术加权投票;不要算出 0.73 这种伪精确分数。3132## 冲突怎么升级3334| 冲突类型 | 处理 | 升级到 |35| --- | --- | --- |36| 事实字段不一致(年份、样本量) | 回查原始来源,以 E1/E2 为准 | 不升级 |37| 纳入/排除判断不一致 | 补一轮检索,确认检索式与时间窗 | survey → lead |38| 统计口径不一致 | 以预注册分析计划为准 | methodologist(可否决) |39| 设计与可证伪性冲突 | 停手,不得边跑边改 | methodologist → pi |40| 数据许可/溯源缺失 | 停手,不得使用该数据 | data-steward → pi |41| 结论方向相反且有同等证据 | 并列报告两种解释与判别实验 | lead → pi |4243升级必须带四样:**分歧点、双方证据 id、已尝试的核对动作、需要谁在什么时限内裁决**。44缺任何一项,上级可以退回。4546## 报告结构4748一份 Campaign 报告固定五段,顺序不要改:49501. **结论**:三句话内。答案是什么、证据强度如何、还有什么没解决。512. **证据表**:一行一条证据,字段固定。523. **分歧与裁决**:哪些结论有多票不一致,怎么处理的。534. **缺口**:明确没做到的部分,以及补上需要什么。545. **可复现信息**:运行 id、数据版本、代码位置、环境与 seed。5556### 证据表字段5758| 字段 | 必填 | 说明 |59| --- | --- | --- |60| `claim` | 是 | 被支持的那句话,一句话 |61| `evidenceId` | 是 | 库内 id / DOI / run id |62…
Excerpt shown under the repository's MIT license.
More skills in Deepscient
- ▸science-agent-scheduling
- ▸science-campaign-design
- ▸science-discipline-methods
- ▸science-experiment-runbook
- ▸science-figure-for-journal
- ▸science-frontier-to-paper
- ▸science-fulltext-hunting
- ▸science-journal-intel
- ▸science-journal-selection
- ▸science-literature-search
- ▸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.
An AI skill that provides design intelligence for building professional UI/UX across multiple platforms.
