菜单
Review a proposed research experiment for identification, controls, measurements, baseline fairness, feasibility, and stopping criteria. Use before external execution; this Skill never runs code or changes an executor project.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/MengkunLiang/DIG-ResearchOS --skill experiment-design-review命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | experiment-design-review |
| description | Review a proposed research experiment for identification, controls, measurements, baseline fairness, feasibility, and stopping criteria. Use before external execution; this Skill never runs code or changes an executor project. |
| tools | ["read_file","write_file","list_files","finish_task"] |
| strict_tools | true |
| model_tier | standard |
| temperature | 0.15 |
| allowed_read_prefixes | ["user_inputs/experiment-design-review/","ideation/","literature/","experiments/","drafts/","_runtime/skill_sessions/"] |
| allowed_write_prefixes | ["ideation/"] |
| outputs_expected | {"design_review":"ideation/skill_experiment_design_review.md","design_matrix":"ideation/skill_experiment_design_matrix.json"} |
| interaction | {"mode":"guided","language":"zh-CN","summary":"在运行实验前审查假设、对照、数据、指标、统计/资源风险和失败判据；不调用外部执行器。","request_required":true,"request_prompt":"请说明要审查的实验目标、最担心的风险、可用资源和是否有必须保留的实验设置。","example_request":"审查此机制验证实验是否能区分 memory carryover 和 task difficulty；给出最小消融、负对照和停止条件。","required_inputs":[{"id":"hypothesis_or_plan","label":"假设或实验计划","description":"至少包括待检验机制、干预、预期结果和已有资源；可使用 hypothesis-compiler 产物。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":80,"example":"ideation/skill_hypothesis_test_plan.md"}],"optional_inputs":[{"id":"evidence_context","label":"文献或已有结果上下文","description":"可选；用于检查基线公平性与已有证据边界，不代表新的实验结果。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":80,"example":"literature/synthesis.md"}],"outputs":[{"id":"design_review","label":"实验设计审查","path":"ideation/skill_experiment_design_review.md","description":"识别风险、混杂、必要对照、指标、可行性、停止与升级决策的可读审查。"},{"id":"design_matrix","label":"实验设计矩阵","path":"ideation/skill_experiment_design_matrix.json","description":"机器可读的 hypothesis-to-measurement、control、risk、kill-criterion 映射。"}]} |
Review the plan as a falsification problem. For every claimed mechanism, specify the intervention, unit, comparison/control, confounds, measurable outcome, expected null/alternative outcome, data provenance requirement, minimum repetitions or uncertainty requirement if known, and kill criterion. Flag under-identification, unobservable latent variables, leakage, weak baselines, metric gaming, subgroup blindness, and resource infeasibility.
Do not manufacture power calculations, data availability, implementation status, or results. State when a proposed study can only show association or ablation evidence.