reasoning-judge
Automates decision-making processes, classification, and anomaly detection based on user-defined thresholds.
Install
mkdir -p .claude/skills/reasoning-judge && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/15253" && unzip -o skill.zip -d .claude/skills/reasoning-judge && rm skill.zipInstalls to .claude/skills/reasoning-judge
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
推理判断技能:二分类判断、多分类判断、阈值判定、置信度评估、异常检测、排序判断。触发场景:当用户提到"判断"、"分类"、"是否"、"评估"、"异常"、"judge"、"classify"、"threshold"、"confidence"、"anomaly"时加载。Key capabilities
- →Extract judgment keywords from user requests
- →Match judgment types: binary, multi-class, threshold, confidence, anomaly, ranking
- →Define quantitative judgment standards (thresholds, ranges, weights)
- →Execute judgment logic based on defined standards
- →Output structured judgment reports with conclusions and confidence
How it works
This skill processes user requests to determine a judgment type, defines quantitative standards based on the type and background data, executes judgment logic, and outputs a structured report with conclusions and confidence. It handles ambiguous inputs by requesting clarification.
Inputs & outputs
When to use reasoning-judge
- →Classify input data
- →Check for anomalies in metrics
- →Perform threshold-based decision making
- →Rank items based on score
About this skill
Reasoning Judge 推理判断技能
阈值参考
详见:阈值判定参考指南
Phase 1:确定判断类型
输入:用户请求、待判断数据
输出:判断类型标识 + 判断目标描述
步骤
- 读取用户请求,提取判断关键词
- 匹配判断类型:
binary:二分类(是/否、真/假、合格/不合格)multi-class:多分类(属于 A/B/C/D 中哪一类)threshold:阈值判定(数值是否超过某个临界值)confidence:置信度评估(某结论的可信程度)anomaly:异常检测(是否偏离正常范围)ranking:排序判断(按优先级/分值对多个选项排序)
- 确认判断目标:明确"对什么进行判断"
判断节点:是否能明确判断类型?
- ✅ 明确 → 进入 Phase 2
- ❌ 模糊 → 失败处理:列出可能的判断类型选项,请求用户确认
Phase 2:定义判断标准
输入:判断类型 + 背景数据
输出:量化判断标准(阈值、范围、权重)
步骤
- 根据判断类型定义标准:
binary:确定分界阈值(如 score ≥ 60 → 合格)multi-class:定义每类的数值区间(如 [90,100]→优秀,[75,90)→良好)threshold:明确临界值和方向(> / < / ≥ / ≤)confidence:定义置信度分级(≥ 0.9 高,[0.7,0.9) 中,< 0.7 低)anomaly:确定正常范围 [μ - kσ, μ + kσ],k 默认为 3ranking:定义评分公式和排序方向(升序/降序)
- 禁止使用模糊表达:不得使用"比较好"、"差不多"、"较高"等非量化描述
- 记录标准来源(用户提供 / 领域默认值 / 统计推算)
判断节点:所有标准是否已量化?
- ✅ 全部量化 → 进入 Phase 3
- ❌ 存在模糊标准 → 失败处理:标记模糊项,请求用户提供具体数值
Phase 3:执行判断逻辑
输入:待判断数据 + 量化判断标准
输出:每个分支的判断结论
步骤
-
将数据代入判断标准,逐条检验
-
决策树执行(每个分支必须有明确出口):
IF binary: IF value >= threshold → 结论: 正例(PASS) ELSE → 结论: 负例(FAIL) IF threshold: IF value > upper_limit → 结论: 超上限(OVER) ELIF value < lower_limit → 结论: 低于下限(UNDER) ELSE → 结论: 正常(NORMAL) IF anomaly: IF |value - μ| > k × σ → 结论: 异常(ANOMALY), 偏差 = (value-μ)/σ ELSE → 结论: 正常(NORMAL) IF multi-class: 按区间匹配 → 输出对应类别标签 IF confidence: 计算置信度 → 输出置信度分级 + 数值 IF ranking: 按评分公式计算 → 输出排序列表 -
记录每个判断步骤的中间值,保证可追溯
判断节点:所有数据是否均落入某个判断分支?
- ✅ 全部匹配 → 进入 Phase 4
- ❌ 边界值或空值 → 失败处理:记录未匹配项,使用默认分支或标记为
UNDEFINED
Phase 4:输出判断结果
输入:Phase 3 的判断结论
输出:结构化判断报告
步骤
- 输出结论(每条数据对应的判断标签)
- 附上置信度(如可计算:基于数据与阈值的距离/符合特征数量)
- 附上判断依据(使用的标准 + 实际数值对比)
- 若为多条数据:以表格形式展示
输出示例:
判断结果:异常 (ANOMALY)
置信度:高(0.95)
判断依据:
- 检测值: 142.3
- 正常范围: [80, 120](μ=100, σ=13.3, k=3)
- 偏差: (142.3-100)/13.3 = 3.18σ > 3σ 阈值
建议:触发人工复核流程
降级策略:当数据不足以支持高置信度判断时:
- 置信度 < 0.6 → 标记为"信息不足,建议补充数据"
- 数据缺失超过 30% → 降级为人工判断,不自动输出结论
When not to use it
- →When the judgment type is ambiguous and cannot be clarified by the user
- →When quantitative standards cannot be defined due to vague input
- →When data is insufficient to support a high-confidence judgment
Limitations
- →Forbidden to use vague expressions like "comparatively good" or "relatively high"
- →When confidence is < 0.6, it is marked as "insufficient information, suggest supplementing data"
- →When data is missing over 30%, it degrades to manual judgment
How it compares
This skill provides a systematic, quantifiable approach to logical reasoning and decision-making, ensuring transparent and traceable judgments unlike subjective or intuitive assessments.
Compared to similar skills
reasoning-judge side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| reasoning-judge (this skill) | 0 | 3mo | No flags | Advanced |
| docetl | 2 | 2mo | Review | Intermediate |
| slm-lab-benchmark | 1 | 5mo | Review | Advanced |
| document-classification | 0 | 2mo | No flags | Intermediate |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
docetl
ucbepic
Build and run LLM-powered data processing pipelines with DocETL. Use when users say "docetl", want to analyze unstructured data, process documents, extract information, or run ETL tasks on text. Helps with data collection, pipeline creation, execution, and optimization.
slm-lab-benchmark
kengz
Run SLM-Lab deep RL benchmarks, monitor dstack jobs, extract results, and update BENCHMARKS.md. Use when asked to run benchmarks, check run status, extract scores, update benchmark tables, or generate plots.
document-classification
leviadi-lang
Use when working on document classification — `classify_document` tool, OCR pipeline, document type taxonomy for insurance/pension documents.
image-analysis
ComeOnOliver
图片分析与识别,可分析本地图片、网络图片、视频、文件。适用于 OCR、物体识别、场景理解等。当用户发送图片或要求分析图片时必须使用此技能。
tss-pipeline
LocNguyen-247
Use when implementing or debugging the TSS remote-sensing workflow in this workspace: Landsat/Sentinel preprocessing, ACOLITE atmospheric correction, cloud/water masking, adjacency correction, station matchup, and model training.
recursive-decision-ledger
yashas-30
Use when the user asks for repeated rollouts, marked decision processes, high-dimensional search, stochastic optimization, local-optima exploration, ensemble comparison, or recursive reasoning with a visible evidence trail.