Token导航 LogoToken导航

Agent Skill · 研究检索

reasoning-verifier(推理验证器)

reasoning-verifier 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

reasoning-verifier 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为只读;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 69,971。

总安装 214周安装 9GitHub Stars 37下载 75核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:reasoning-verifier(推理验证器)
来源仓库:https://github.com/majesticlabs-dev/majestic-marketplace
仓库路径:skills/reasoning-verifier
安装命令:
npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill reasoning-verifier
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill reasoning-verifier

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

reasoning-verifier 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。

安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。当前顶部介绍为空,原始 SKILL.md 摘录未提供。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

来源仓库内路径:skills/reasoning-verifier

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

reasoning-verifier 怎么安装?

在终端执行 npx skills add https://github.com/majesticlabs-dev/majestic-marketplace --skill reasoning-…。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

reasoning-verifier 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

reasoning-verifier 需要什么权限?

权限级别是只读。该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

reasoning-verifier 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

reasoning-verifier 用的人多吗?

公开数据是累计安装 214 次,近一周安装 9 次,下载 75 次,GitHub 37 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

reasoning-verifier 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 只想先试试效果、不希望它改动本地文件的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
reasoning-verifier

推理验证器 · 本页

21437只读未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

Reasoning Verifier

Verify reasoning using the Reverse Chain-of-Thought (RCoT) methodology to detect and correct errors in LLM-generated solutions by systematically comparing what the problem stated versus what the solution assumed.

When to Use

  • After complex multi-step reasoning or analysis tasks
  • When solution correctness is critical
  • To catch overlooked conditions from the original problem
  • To identify hallucinated assumptions not present in the original problem
  • Before finalizing recommendations or decisions based on LLM analysis

RCoT Methodology

Follow these steps in order:

Step 1: Problem Decomposition

Extract all conditions from the original problem:

  1. Read the original problem statement carefully
  2. List every explicit condition, constraint, and requirement
  3. Note any implicit conditions that can be logically inferred
  4. Number each condition for reference

Format:

## Original Conditions

1. [Condition from original problem]
2. [Condition from original problem]
3. [Condition from original problem]
...

Step 2: Solution Reconstruction

Analyze the provided solution and reconstruct what problem it appears to solve:

  1. Read through the solution's reasoning steps
  2. Identify every assumption the solution made
  3. List conditions the solution used in its reasoning
  4. Note the final answer/conclusion

Format:

## Reconstructed Conditions (from solution)

1. [Condition the solution assumed]
2. [Condition the solution assumed]
3. [Condition the solution assumed]
...

## Solution's Answer: [final answer]

Step 3: Condition Comparison

Compare the two condition lists to find discrepancies:

Overlooked Conditions: Present in original but NOT used in solution

  • These are facts the solution ignored
  • May cause incorrect answers if they affected the logic

Hallucinated Conditions: Present in solution but NOT in original

  • These are assumptions the solution invented
  • Invalid if they cannot be logically deduced from original conditions

For each candidate condition, ask: "Can this be logically deduced from the other condition list?"

Format:

## Comparison Results

### Overlooked Conditions
- [Condition X] from original was not addressed
  - Impact: [How this affects the solution]

### Hallucinated Conditions
- [Condition Y] was assumed but not stated
  - Validity: [Can this be deduced? YES/NO]
  - Impact: [How this affects the solution]

Step 4: Error Classification

Classify the severity of each discrepancy:

  • Critical: Changes the final answer
  • Major: Affects intermediate reasoning significantly
  • Minor: Technical oversight with no impact on answer
  • None: Solution is correct

Step 5: Revision Prompt Generation

If errors are found, generate targeted correction prompts:

For overlooked conditions:

You have ignored some real conditions:
1. [overlooked condition]
Here are detailed reasons:
1. [Why this condition matters and how it affects the answer]

For hallucinated conditions:

You have assumed conditions not in the problem:
1. [hallucinated condition]
This cannot be logically derived because:
1. [Why this assumption is invalid]

Step 6: Revised Solution

If errors were found, provide the corrected solution:

  1. Acknowledge the specific mistakes
  2. Incorporate all overlooked conditions
  3. Remove invalid hallucinated assumptions
  4. Show corrected reasoning step-by-step
  5. Provide the revised answer

Output Format

# RCoT Verification Report

## Problem Summary
[Brief description of the problem being verified]

## Original Conditions
1. [condition]
2. [condition]
...

## Reconstructed Conditions (from solution)
1. [condition]
2. [condition]
...

## Comparison Results

### Overlooked Conditions
[List or "None found"]

### Hallucinated Conditions
[List or "None found"]

## Verdict: [CORRECT / NEEDS REVISION]

## Severity: [Critical / Major / Minor / None]

---

[If NEEDS REVISION:]

## Revision Prompt

[Generated prompt to correct the solution]

## Revised Solution

[Corrected answer with proper reasoning]

Example

Original Problem: "Mary has 40 window ledges. She has 2 potted plants on each ledge. Yesterday, she received 18 new potted plants from her favorite nursery. She decided to give away 1 potted plant from each ledge. How many potted plants will Mary remain with?"

Original Solution: "Mary has 2 x 40 = 80 plants. After giving away 1 from each ledge, she has 80 - 40 = 40 plants."

RCoT Analysis:

  • Overlooked: "She received 18 new potted plants" - this condition was completely ignored
  • The solution is INCORRECT

Revised Solution: "Mary starts with 2 x 40 = 80 plants. She receives 18 new plants, giving her 80 + 18 = 98 plants. After giving away 1 from each of 40 ledges, she has 98 - 40 = 58 plants."

平台分布

Codex

38.98%
按下载量换算29

Claude

29.98%
按下载量换算22

Cursor

17.82%
按下载量换算13

Gemini CLI

9.97%
按下载量换算7

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills