research-lab
Skill4 starsChanged 21 days agoMITView on GitHub
使用独立 Research Core 把模型比较、软件评估或其他可证伪问题转化为可恢复、可复现、证据可追溯的研究。 仅用于需要真实执行对照实验的场景:模型/方案/系统的实证对比、基线比较、模型替代评估、消融、盲评、可复现实验。 不用于简单事实查询、只读审计/盘点/调查(即使涉及"证据")、无需实验的低风险判断,或用户只要求普通文案/解释的场景。
3 other projects use a version of this.
Adopted by
Most recently taken up by
- ooooooooooooooooooop/personal-aiword for word, first seen Oct 22, 2026
- BigY0shi/super-skillstheir own version, first seen May 7, 2026
- Xplore-LAB/openclaw-scholar-packtheir own version, first seen Apr 2, 2026
Other versions3 projects use a version of this.
Word for word · 1
- ooooooooooooooooooop/personal-ai★ 4100% same
Edited · 0
None found.
Wrote their own version · 2
- BigY0shi/super-skills★ 4same name
- Xplore-LAB/openclaw-scholar-pack★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository23 other files it gives its agents.
- AGENTS.mdAGENTS.md
- dsh-config/AGENTS.mdAGENTS.md
- chatgpt-web-bridgeSkill
- clarify-before-changeSkill
- decision-gatesSkill
- dsh-config-syncSkill
- environment-bootstrapSkill
- evolution-proposalSkill
- execution-disciplineSkill
- find-sessionSkill
- minimal-implementationSkill
- natural-rewriteSkill
- personal-ai-operations-reviewSkill
- publish-and-reuseSkill
- simulate-elite-expertsSkill
- skill-quality-gateSkill
- skill-repository-maintainerSkill
- subagent-execution-governanceSkill
- systematic-optimizationSkill
- task-mode-routerSkill
- unified-taskflowSkill
- world-model-runtimeSkill
- your-skill-nameSkill
The file
Files
Research Lab
目标
Research Skill 是独立 Research Runtime 的使用协议,不是研究系统本体。它负责识别研究任务、约束研究设计、调用稳定 research_ 前缀工具,并依据 Evidence 返回结论;实验、统计、Workspace、同步和恢复由 Research Core 负责。
触发门禁
满足任一条件时使用:
- 用户要求比较模型、方案或系统,且需要质量、成本或延迟证据。
- 用户提出“能否替代”“是否显著更好”“哪个更可靠”等可证伪问题。
- 结论依赖数据集、Baseline、重复运行、盲评、消融或环境记录。
- 用户要求恢复、复现、审计或跨设备继续既有 Research。
不要用于简单事实查询、单一确定答案、无需实证的文案建议或低成本可逆选择。
硬规则
- 不得从用户问题直接跳到结论;先建立或验证 ResearchSpec。
- Spec 至少明确 question、hypotheses、单个
baseline、candidates、dataset/cases、metrics、decision rules 和 adapters;执行字段以当前research validate为准。 - 不具备所需模型、数据或执行能力时返回
UNSUPPORTED,不得模拟实验结果。 - Spec 或执行过程无效时返回
INVALID;有效实验但证据不足时返回INCONCLUSIVE。 - 结论必须引用 Evidence、Run 和 Artifact digest;不能只复述模型输出。
- 不把 API key、token、私钥、绝对用户路径或 credential value 放入 Spec、Workspace、CAS 或工具参数;只使用设备本地配置或 credential reference。
- 多设备 revision 冲突必须显式呈现,禁止静默选择最后写入版本。
标准流程
- 识别任务:把问题改写成可证伪 question;明确候选方案与 Baseline。
- 建立 Spec:若用户已有 Spec,调用
research_validate;否则生成便携 JSON 文件后调用research_create。 - 执行前检查:用
research_inspect检查 revision、数据集、Adapter 和决策阈值。缺关键条件时停止执行并返回结构化缺口。 - 执行或恢复:调用
research_execute。已有 run 时传入 run id;中断后调用research_continue,不要创建重复研究冒充恢复。 - 检查状态:用
research_status查看已完成/失败/待执行项,不进行热轮询;等待真实执行反馈。 - 比较与证据:调用
research_compare,再用research_evidence获取支撑与限制。 - 返回结论:只使用 Runtime 的 Decision 枚举:
SUPPORTED、REJECTED、INCONCLUSIVE、UNSUPPORTED、INVALID。
LLM 对比最小 Spec
{
"protocolVersion": "1.0",
"researchId": "model-replacement-evaluation",
"question": "Candidate 是否可替代 Baseline 完成指定任务?",
"taskType": "llm-comparison",
"hypotheses": [{"id": "h1", "statement": "candidate 质量不低于 baseline 且成本更低"}],
"baseline": {"id": "baseline", "adapter": "command"},
"candidates": [{"id": "candidate", "adapter": "command"}],
"dataset": {
"cases": [{"id": "case-1", "input": "...", "expected": "..."}]
},
"metrics": ["exact_match", "latency_ms", "cost"],
"decisionRules": {
"minDelta": 0.0,
"preferred": "candidate"
},
"adapters": {
"command": {
"type": "command",
"command": ["python", "path/to/adapter.py"]
}
}
}字段以安装版本的 research validate 为准;不要凭 Skill 正文猜测 Core 已新增的字段。
输出契约
最终报告包含:
Research ID / revision
Question 与 Hypothesis
Baseline、Candidate、数据集与指标
Run 状态和环境/Adapter 版本
Decision
关键 Evidence(含 artifact digest)
限制、冲突与尚缺证据
复现或继续命令不得把 ok: true、工具成功或单次模型输出当作研究结论。相关示例见 examples/llm-replacement.md。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

