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 compare-pair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。
compare-pair
Purpose
Execute one pairwise comparison under a declared criterion and return a winner, confidence, tie status, and reasons.
Input contract
yaml
required: [item_a, item_b, comparison_criterion, comparison_model, evidence_register]
optional: [tie_rule, prior_ratings]
constraints: [criterion direction and evidence scope are fixed before comparison]Procedure
- Verify that both items are comparable under the same criterion and evidence scope.
- Assess item A and item B symmetrically, recording criterion-level evidence and limitations.
- Select A, B, or tie; assign confidence on the declared scale and explain the decisive contrast.
- Return the comparison record without updating global ratings.
Output contract
yaml
produces: [pairwise_result, confidence, rationale, evidence_gaps]
delta_fields: [findings, evidence_updates, uncertainties, decisions]Quality gates
- The result includes exactly one winner or an explicit tie, confidence, criterion, and symmetric evidence notes.
- Confidence cannot exceed the support of the weaker side of the comparison.
Failure and counterexamples
Do not compare items with different scopes or criteria. If evidence is asymmetric or incomparable, return unresolved rather than forcing a winner.
Provenance map
- resolved: comparison-executor
- resolved: gap-pairwise-judgment
