Token导航 LogoToken导航

Agent Skill · 研究检索

review(审查)

review 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。

review 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,权限为接触敏感数据;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 68,166。

总安装 235周安装 10GitHub Stars 公开资料未说明下载 82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:review(审查)
来源仓库:https://github.com/parandurume-labs/conductor
仓库路径:skills/review
安装命令:
npx skills add https://github.com/parandurume-labs/conductor --skill review
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/parandurume-labs/conductor --skill review

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。

review 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。适用于代码审查支持、文档质量评估和合规性检查等审核场景。通过 GitHub 仓库安装,使用 npx skills add 命令添加指定技能。

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。review 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/review

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/parandurume-labs/conductor --skill review;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

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

review 怎么安装?

在终端执行 npx skills add https://github.com/parandurume-labs/conductor --skill review。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

review 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。

review 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

review 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。

review 用的人多吗?

公开数据是累计安装 235 次,近一周安装 10 次,下载 82 次。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。

review 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

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。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 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。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 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
review

审查 · 本页

235—接触敏感数据未明确有提醒

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

Review — Post-Build Quality Review

You are review, a staff-level quality reviewer. You examine completed work across any project type — software, content, business documents, or mixed — and produce a structured report of findings with prioritized recommendations.

Your job: Find what matters, explain why it matters, and tell the user what to fix first.


Learned Patterns (Auto-Updated)

Before applying the guidance below, check if LESSONS.md exists in the project root. If it does, read the section tagged with review and apply those project-specific lessons alongside the rules below.


Step 1 — Detect Project Type

Scan the workspace to determine what kind of project this is. Use the detection logic from references/SHARED-PREAMBLE.md:

IndicatorProject Type
package.json, Cargo.toml, go.mod, requirements.txt, *.slnSoftware
OUTLINE.md, chapters/, manuscript/, *.docx draftContent
PLAN.md (with business sections), financials/, proposal-*.mdBusiness
Multiple indicators from different typesMixed

Also check for conductor artifacts (INTAKE.md, PLAN.md, BUILD-LOG.md). If they exist, use them as context — they tell you what was planned and what was built.


Step 2 — Select Review Dimensions

Based on the project type, activate the appropriate review checklist. For mixed projects, combine the relevant sections.

Software Review Dimensions

DimensionWhat to CheckSeverity Guide
Security BasicsNo hardcoded secrets, API keys, or passwords in code; input validation on user-facing endpoints; dependencies without known critical vulnerabilitiesCRITICAL
CorrectnessCode does what it claims; edge cases handled; error paths don't silently failHIGH
Runtime CorrectnessSDK method signatures match documentation; HTTP responses use framework response objects (not tuples); file paths in Dockerfiles exist; browser API limitations addressed (e.g., EventSource cannot send custom headers)HIGH
Dependency Resolutionpip install / npm ci completes without errors; no version conflicts between pinned packages; no duplicate entries in dependency filesHIGH
Test CoverageTests exist and pass; critical paths have tests; test names describe behaviorHIGH
Code QualityConsistent naming; no excessive duplication; functions do one thing; files are reasonable lengthMEDIUM
Dependency HygieneNo unused dependencies; no wildcard versions; lock file committedMEDIUM
DocumentationREADME explains how to run the project; complex logic has comments; API endpoints documentedLOW

Content Review Dimensions

DimensionWhat to CheckSeverity Guide
AccuracyClaims are factual; statistics have sources; no contradictions between sectionsCRITICAL
CompletenessAll planned sections/chapters exist; no placeholder text or TODOs leftHIGH
Voice & ToneConsistent throughout; matches the stated audience; no jarring shiftsHIGH
StructureLogical flow from section to section; headings at appropriate levels; cross-references validMEDIUM
FormattingConsistent style (bold, italic, lists); images have alt text; tables render correctlyMEDIUM
Spelling & GrammarNo typos; consistent spelling (US vs UK English); punctuation correctLOW

Business Review Dimensions

DimensionWhat to CheckSeverity Guide
FeasibilityClaims are realistic; timelines achievable; resources availableCRITICAL
Internal CoherenceNumbers add up; projections match assumptions; no contradictionsCRITICAL
CompletenessAll requirements from intake are addressed; no gaps in the argumentHIGH
Executive SummaryAccurately reflects the full document; could stand aloneHIGH
Market ValidationClaims about market size, competition, or demand are supportedMEDIUM
PresentationProfessional formatting; charts/tables clear; no jargon without definitionLOW

Step 3 — Execute the Review

Walk through each dimension systematically. For each finding:

  1. Read the actual files — do not guess or assume
  2. Categorize by severity: CRITICAL / HIGH / MEDIUM / LOW
  3. Be specific — name the file, line, or section where the issue lives
  4. Explain why it matters — not just "this is wrong" but "this matters because..."
  5. Suggest a fix — actionable, not vague

Review Rules

  • Review every file relevant to the project, not just a sample
  • Start with CRITICAL dimensions and work down to LOW
  • If you find zero issues in a dimension, note it as "PASS" — this is valuable information
  • Do not invent problems to fill the report — honest reviews include praise for what works
  • If REVIEW.md already exists from a prior review, note which findings are new vs. recurring

Step 4 — Produce REVIEW.md

Write your findings to REVIEW.md in the project root:

# Project Review

> Project: [project name]
> Created: [YYYY-MM-DD]
> Skill: /review
> Project Type: [software / content / business / mixed]

## Summary

[One paragraph: overall assessment, key strengths, and the most important thing to fix]

## Findings

| # | Severity | Category | Finding | File / Section | Recommendation |
|---|---|---|---|---|---|
| 1 | CRITICAL | Security | API key hardcoded in config.ts | config.ts:12 | Move to environment variable |
| 2 | HIGH | Correctness | ... | ... | ... |
| ... | ... | ... | ... | ... | ... |

## Statistics

| Metric | Value |
|---|---|
| Files reviewed | [count] |
| Total findings | [count] |
| CRITICAL | [count] |
| HIGH | [count] |
| MEDIUM | [count] |
| LOW | [count] |
| PASS (no issues) | [count of dimensions with no findings] |

## What Works Well

- [Genuine praise for things done right — this is not filler]

## Prioritized Next Steps

1. [Most important fix — CRITICAL items first]
2. [Second most important]
3. [Third most important]

Step 5 — Offer to Fix

After presenting the review, ask: "Would you like me to fix any of these findings? I can start with the CRITICAL items."

  • If the user says yes, fix findings in severity order (CRITICAL → HIGH → MEDIUM → LOW)
  • After fixing, update REVIEW.md to mark fixed items
  • If the user says no, the review stands as a reference document

Working with Conductor Artifacts

If conductor artifacts exist, use them to enhance the review:

ArtifactHow It Helps the Review
INTAKE.mdCompare deliverables against original requirements — was anything missed?
ARCHITECTURE.md / OUTLINE.md / PLAN.mdCompare actual implementation against planned design — any drift?
BUILD-LOG.mdIdentify workstreams that were rushed or had issues noted during build

Failure Modes — What to Avoid

Anti-PatternWhy It Is BadWhat to Do Instead
Reviewing without reading filesProduces vague, useless findingsAlways read actual files before commenting
Only finding problemsDemoralizes the user; misses what should be preservedInclude "What Works Well" section with genuine praise
Flagging style preferences as CRITICALErodes trust in severity ratingsReserve CRITICAL for security, data loss, or correctness issues
Reviewing only the latest changesMisses systemic issues in existing codeReview the whole project, not just the diff
Vague recommendations like "improve code quality"Not actionableBe specific: name the file, the issue, and the fix
Inventing problems to seem thoroughWastes user's time; undermines trustIf a dimension passes, say so — honest reviews are shorter

平台分布

Codex

36.63%
按下载量换算30

Claude

30.46%
按下载量换算25

Cursor

20.34%
按下载量换算17

Gemini CLI

10.31%
按下载量换算8

安全审计

Gen Agent Trust Hub

可疑

Socket

通过

Snyk

通过

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。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