
Code Review Council
- 27 installs
- 4 repo stars
- Updated June 2, 2026
- adjfks/corner-skills
code-review-council is a Claude Code skill that runs a multi-role code review committee of five independent reviewer agents that analyze code in parallel and cross-check each other's conclusions.
About
code-review-council runs a multi-role code review as five independent reviewers (Explorer, Analyst, Architect, Librarian, Critic) that work in parallel, then a Critic cross-checks their conclusions into one combined report. A developer uses it to understand unfamiliar code, diagnose bugs or performance issues, weigh implementation tradeoffs, or vet a library choice. The design breaks the confirmation bias of a single linear analysis. Documentation is in Chinese.
- Multi-agent code review with five independent reviewer roles
- Explorer, Analyst, Architect, Librarian, and Critic work in parallel then cross-check
- Aims to break single-thread confirmation bias in code analysis
Code Review Council by the numbers
- 27 all-time installs (skills.sh)
- Ranked #686 of 1,352 Code Review & Quality skills by installs in the Skillselion catalog
- Data as of Jul 28, 2026 (Skillselion catalog sync)
code-review-council capabilities & compatibility
- Capabilities
- code review · bug triage · architecture review · dependency selection
- Use cases
- code review · debugging
- Pricing
- Free
What code-review-council says it does
Code Review Council 是一个多角色审查框架。
**核心价值**:打破单线程分析的"确认偏误"——先找到方案就锚定判断的倾向。
npx skills add https://github.com/adjfks/corner-skills --skill code-review-councilAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 27 |
|---|---|
| repo stars | ★ 4 |
| Last updated | June 2, 2026 |
| Repository | adjfks/corner-skills ↗ |
What it does
Run a five-role parallel code review that cross-checks conclusions for bug hunts, design tradeoffs, and library selection.
Who is it for?
Complex bug hunts, implementation-design decisions, and library selection where a single analysis risks confirmation bias.
When should I use this skill?
A user asks to review code, diagnose a bug, decide how to implement a feature, or pick a library.
What you get
A combined report where five reviewer roles' findings are cross-checked by a Critic.
- code map
- ranked problem list with root causes
- solution comparison table
By the numbers
- 5 reviewer roles
- Critic must find at least 2 problems in others' work
- Architect must give at least 2 solution paths
Files
Code Review Council
概述
Code Review Council 是一个多角色审查框架。面对代码问题时,不是由单个 Claude 顺序完成所有步骤,而是召集五个独立视角的「委员」并行工作,最终由审查者交叉验证所有结论,输出一份综合报告。
核心价值:打破单线程分析的"确认偏误"——先找到方案就锚定判断的倾向。
---
五位委员角色
1. Explorer(探索者)
职责:广度优先,描述现状,不提方案。
- 梳理代码库结构、模块边界、文件组织
- 追踪关键调用链和数据流向
- 标记入口点、副作用、外部依赖
- 输出:代码地图 + 关键路径清单
约束:Explorer 只描述"是什么",禁止说"应该怎么做"。
---
2. Analyst(分析者)
职责:深度挖掘问题根因,独立于 Explorer 得出结论。
- 分析性能瓶颈(时间复杂度、内存、IO)
- 识别潜在 bug、边界条件、竞态条件
- 评估技术债务和风险点
- 输出:问题清单(按严重程度排序)+ 根因分析
约束:Analyst 必须给出根因,不接受"可能是 X 导致的"这种模糊结论。
---
3. Architect(方案者)
职责:独立提出 2-3 个实现方案,明确每个方案的 tradeoff。
- 每个方案必须包含:核心思路、实现复杂度、性能特征、可维护性
- 标注方案的前提假设
- 给出推荐方案及理由
- 输出:方案对比表 + 推荐方案的伪代码或关键实现片段
约束:Architect 不能只提一个方案,必须覆盖至少两种不同路径。
---
4. Librarian(库研究员)
职责:调研与任务相关的开源库,验证实际可用性。
- 确认库的当前版本、维护状态(是否活跃)
- 验证关键 API 的实际用法(不能凭记忆,要基于文档)
- 列出已知坑、breaking change、版本兼容性问题
- 对比同类库的优劣
- 输出:库选型建议 + 关键 API 示例代码
约束:Librarian 必须区分"确认过的"和"印象中的"信息,不能混淆。
---
5. Critic(审查者)⚡
职责:横向审查其他四位委员的结论,找矛盾、遗漏和错误假设。这是最后一个发言的角色。
- 验证 Explorer 的代码地图是否遗漏关键路径
- 挑战 Analyst 的根因是否真的是根因
- 质疑 Architect 方案的隐含假设是否成立
- 核查 Librarian 推荐的库是否真的适用于当前场景
- 输出:审查意见清单 + 最终综合建议
约束:Critic 必须至少找出 2 处其他委员的问题或遗漏,不能全部认可。
---
执行流程
第一轮:并行工作
├── Explorer → 代码地图
├── Analyst → 问题清单
├── Architect → 方案对比
└── Librarian → 库选型
第二轮:交叉审查
└── Critic 阅读所有输出 → 审查意见
第三轮:综合输出
└── 整合所有结论 → 最终建议---
输出格式规范
每位委员的输出块用以下格式包裹:
## [角色名] 的报告
### 结论
(核心发现,3-5 条)
### 详细分析
(具体内容)
### 置信度
高 / 中 / 低(+ 说明原因)Critic 额外输出:
## [Critic] 的审查意见
### 发现的问题
1. 针对 Explorer:...
2. 针对 Analyst:...
3. 针对 Architect:...
4. 针对 Librarian:...
### 综合建议
(整合所有视角后的最终建议)---
使用指南
何时启用所有五位委员
- 复杂的 bug 排查
- 重要功能的方案设计
- 技术选型决策
何时只启用部分委员
根据任务类型裁剪:
| 任务类型 | 启用角色 |
|---|---|
| 理解陌生代码 | Explorer + Analyst + Critic |
| 方案设计 | Architect + Analyst + Critic |
| 库选型 | Librarian + Critic |
| Bug 排查 | Explorer + Analyst + Critic |
| 全面评审 | 全部五位 |
执行原则
1. 各委员独立:后发言的委员不应预判前面委员的结论 2. Critic 最后:必须等其他委员完成后再发言 3. 置信度透明:不确定的结论必须标注置信度 4. 聚焦用户问题:所有分析服务于用户的原始问题,不做无关展开
---
示例触发语
- "帮我分析一下这段代码有什么问题"
- "我要实现 X,用哪种方案好"
- "这个库怎么用,有没有坑"
- "帮我看看这个 PR"
- "为什么这里会内存泄漏"
- "重构这个模块,怎么设计"
Related skills
FAQ
What are the five roles?
Explorer (code map), Analyst (root-cause problems), Architect (2-3 solution options), Librarian (library research), and Critic (cross-checks the other four).
Do all five roles always run?
No. The skill trims roles by task type, e.g. Librarian plus Critic for library selection, Explorer plus Analyst plus Critic for bug hunts.