跳到主内容
智客 ZICQ

技能库 智客分类:Agent 工作流 review

审查

集中个人驱动的代码审查(Fowler,Torvalds,Carmack,React Core等). 因此克洛德可以在用户要求代码审查或针对特定视角的批评时选择或综合专家观点.

37 安装量

官方网址:skills.sh

技能介绍

先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。

做什么

集中个人驱动的代码审查(Fowler,Torvalds,Carmack,React Core等). 因此克洛德可以在用户要求代码审查或针对特定视角的批评时选择或综合专家观点.

何时用

官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。

代理如何加载

按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Review Skill、Critical Workflow、General Checklist、Multi-Perspective Reviews、Persona References (load on demand)。 其中含规范建议的小节:分步指令。

文件分析

文件分析:除 SKILL.md 外,正文引用了 references/ai-reviewer.md、references/anders-reviewer.md、references/beck-reviewer.md、references/bjarne-reviewer.md、references/brendan-reviewer.md、references/carmack-reviewer.md,属于带资源的技能包,这些文件按需再读。

官方 description(原文)

Centralizes persona-driven code reviews (Fowler, Torvalds, Carmack, React core, etc.) so Claude can pick or combine expert viewpoints when the user asks for a code review or perspective-specific critique.

Review SkillCritical WorkflowGeneral ChecklistMulti-Perspective ReviewsPersona References (load on demand)

· allowed-tools:Read, Grep, Glob, Bash

来源分类:skills.sh agent-skill

SKILL.md 与 Agent 调用

官方规范 ↗
name
review
description
Centralizes persona-driven code reviews (Fowler, Torvalds, Carmack, React core, etc.) so Claude can pick or combine expert viewpoints when the user asks for a code review or perspective-specific critique.
allowed-tools
Read, Grep, Glob, Bash实验字段,支持情况取决于客户端;字段声明本身不会授予工具权限。
  1. 发现技能客户端向 Agent 提供名称与描述目录。
  2. 匹配与调用用户指定或任务匹配后,载入 SKILL.md 指令。
  3. 按需加载按步骤读取参考文档、使用脚本与素材。
指令中引用的文件 · 12
  • references/ai-reviewer.md
  • references/anders-reviewer.md
  • references/beck-reviewer.md
  • references/bjarne-reviewer.md
  • references/brendan-reviewer.md
  • references/carmack-reviewer.md
  • references/dean-reviewer.md
  • references/dhh-reviewer.md
  • references/fowler-reviewer.md
  • references/github-reviewer.md
  • references/grace-reviewer.md
  • references/guido-reviewer.md

以下路径提取自原文;文件是否齐全请以来源仓库中的完整目录为准。

具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗

安装这个技能

Skills CLI ↗

先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。

该技能引用了附属文件,请从来源获取完整目录;仅复制 SKILL.md 可能缺少依赖。

交给 Agent 安装

复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。

把 Agent Skill「review」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-f7db3d7908f3efdd-%E5%AE%A1%E6%9F%A5.html
请存为 .cursor/skills/review/SKILL.md 或 .claude/skills/review/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
该技能还带 scripts/、references/、assets/ 等文件,请从 https://github.com/johnlindquist/claude 取完整目录,不要只建一个 SKILL.md。

GitHub 完整包 ↗

终端安装 · Skills CLI

需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。

npx skills add 'https://github.com/johnlindquist/claude' --list

npx skills add 'https://github.com/johnlindquist/claude' --skill 'review'

CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。

阅读排版

name: review description: Centralizes persona-driven code reviews (Fowler, Torvalds, Carmack, React core, etc.) so Claude can pick or combine expert viewpoints when the user asks for a code review or perspective-specific critique. allowed-tools: Read, Grep, Glob, Bash

Review Skill

Unifies every reviewer persona into one Skill. Claude activates this Skill whenever code should be reviewed and then "lazy loads" the exact perspective by opening the reference docs linked below or by spawning persona-specific subagents.

Critical Workflow

REQUIRED: Before conducting ANY code review, you MUST load the relevant persona reference file(s) using the Read tool. These references contain the specific review priorities, perspective, and evaluation criteria for each reviewer persona.

  1. Collect the code/diff context plus the user's goals (bugs, architecture, performance, etc.).
  2. MANDATORY: Parse reviewer hints (e.g., "perf, react, typescript") and READ the matching reference file(s) directly using the Read tool BEFORE reviewing:
    • AI/ML concerns → Read references/ai-reviewer.md FIRST
    • Type system concerns → Read references/anders-reviewer.md FIRST
    • Testing/TDD concerns → Read references/beck-reviewer.md FIRST
    • Performance/abstraction → Read references/bjarne-reviewer.md FIRST
    • Innovation/pragmatism → Read references/brendan-reviewer.md FIRST
    • Low-level performance → Read references/carmack-reviewer.md FIRST
    • Distributed systems → Read references/dean-reviewer.md FIRST
    • Convention/simplicity → Read references/dhh-reviewer.md FIRST
    • Refactoring/architecture → Read references/fowler-reviewer.md FIRST
    • Collaboration/CI/CD → Read references/github-reviewer.md FIRST
    • Abstraction/modularity → Read references/grace-reviewer.md FIRST
    • Readability/Python → Read references/guido-reviewer.md FIRST
    • Portability/Java → Read references/james-reviewer.md FIRST
    • Compiler/tooling → Read references/lattner-reviewer.md FIRST
    • Systems/rigor → Read references/linus-reviewer.md FIRST
    • Developer joy/Ruby → Read references/matz-reviewer.md FIRST
    • Observability/tracing → Read references/perf-reviewer.md FIRST
    • React patterns → Read references/react-reviewer.md FIRST
    • Go/concurrency → Read references/rob-reviewer.md FIRST
    • Unix philosophy → Read references/unix-reviewer.md FIRST
  3. Apply the reviewer persona's perspective by following their specific guidance and priorities from the loaded reference.
  4. Cite specific files/lines, flag issues, and provide concrete recommendations.

DO NOT attempt to conduct a code review without first loading the appropriate persona reference file(s).

General Checklist

  • Understand inputs/outputs, dependencies, and expected behavior before judging the change.
  • Use the allowed tools (Read, Grep, Glob, Bash) to inspect implementation, history, and tests.
  • Evaluate correctness, safety, performance, maintainability, and user impact.
  • Flag missing tests, weak docs, regressions, or architectural drift; propose concrete fixes.
  • Summarize findings in severity order, then note risks, questions, and verification steps.

Multi-Perspective Reviews

When multiple personas are requested (e.g., "review this with Anders and React perspectives"):

  • Read each relevant reference file from the list below
  • Apply each perspective's priorities and concerns to the code
  • Synthesize findings: highlight where perspectives agree or conflict
  • Prioritize issues by severity across all perspectives

Persona References (load on demand)

  • AI Visionaries - adaptive systems, emergent behavior, data-driven design. Open instructions
  • Anders Hejlsberg - strong typing, language/tooling ergonomics, structured APIs. Open instructions
  • Kent Beck - TDD discipline, rapid feedback loops, adaptive design. Open instructions
  • Bjarne Stroustrup - performance via abstraction, type safety, disciplined engineering. Open instructions
  • Brendan Eich - rapid innovation, creative problem-solving, pragmatic experimentation. Open instructions
  • John Carmack - low-level excellence, graphics/perf tuning, precision thinking. Open instructions
  • Jeff Dean - planet-scale systems, efficiency, distributed reliability. Open instructions
  • DHH - opinionated conventions, developer autonomy, simplicity over ceremony. Open instructions
  • Martin Fowler - refactoring readiness, evolutionary architecture, intentional design. Open instructions
  • GitHub Generation - collaboration hygiene, docs, CI/CD automation. Open instructions
  • Grace Hopper & Barbara Liskov - abstraction integrity, substitutability, modular design. Open instructions
  • Guido van Rossum - readability, Pythonic simplicity, pragmatic clarity. Open instructions
  • James Gosling - JVM portability, API stability, backward compatibility. Open instructions
  • Chris Lattner - compiler/toolchain innovation, language interoperability, performance. Open instructions
  • Linus Torvalds - kernel-level rigor, patch discipline, brutally honest feedback. Open instructions
  • Yukihiro "Matz" Matsumoto - Ruby aesthetics, human-centric design, joy in code. Open instructions
  • Brendan Gregg & Liz Rice - observability, tracing, data-first performance analysis. Open instructions
  • React Core Maintainer - hooks, concurrent rendering, DX-focused component patterns. Open instructions
  • Rob Pike - Go/Unix minimalism, concurrency primitives, composable tooling. Open instructions
  • Unix Traditionalist - small sharp tools, composability, text-first automation. Open instructions

Each reference stays out of context until explicitly opened, keeping Claude's context lean while still giving fast access to the original, detailed reviewer guidance.

相关技能

Agent 工作流

技能创建者Skill Creator

创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.

Agent 工作流

克劳德胡布Clawdhub

使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.

Agent 工作流

团队指挥Agent Team Orchestration

管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.

Agent 工作流

超级力量Superpowers

Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…