使用方法
- 复制安装命令，在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物，并对照来源检查输出。
npx skills add https://github.com/yogsoth-ai/de-anthropocentric-research-engine --skill design-falsification-test开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况，不等于独立质量或安全测试。
仓库原始内容
按源仓库内容呈现，保留标题、案例、代码、表格、链接以及原文引用的演示图片。
design-falsification-test
Purpose
Specify a concrete observation, computation, intervention, counterexample, or measurement that could count against a claim.
Input contract
yaml
required: [sharp_claim, scope, candidate_observables]
optional: [cost, available_tools, competing_predictions]
constraints: [test must identify a failure condition and measurement rule]Procedure
- Enumerate claim commitments and possible failure observations.
- Rank tests by information value and cost.
- Specify the cheapest decisive test with pass/fail interpretation.
If the falsification test has an executable intervention and verdict rule, consider execute-probe as the next tactic.
Output contract
yaml
produces: [falsification_test, failure_condition, measurement_plan, priority_rationale]
delta_fields: [findings, evidence_updates, uncertainties, decisions]Quality gates
- Test must be capable of producing evidence against the claim, not only confirming examples.
Failure and counterexamples
Do not design a test whose acceptance criterion is copied from the claim.
Provenance map
- resolved: red-team-truthseeking
