Token导航 LogoToken导航TokenDH.com
研究检索external-servicegithub未标认证来源可访问clear审计通过

code-review代码审查

Agent Skill

code-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。

总安装

912

周安装

38

GitHub Stars

公开资料未说明

下载量

304
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

3

许可证

MIT

最后核验

2026-05-01

来源状态

来源可访问

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:code-review(代码审查)
来源仓库:https://github.com/carvalab/k-skills
仓库路径:skills/code-review
安装命令:
npx skills add https://github.com/carvalab/k-skills --skill code-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。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/carvalab/k-skills --skill code-review

简介

code-review 执行高级别代码审查,聚焦复杂逻辑与架构问题。

  • 结合 Kavak 特定模式与 CI 模板进行合规检查。
  • 利用 MCP 工具验证企业内部最佳实践。
  • 需接入 kavak-platform 文档系统获取最新标准。
  • 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。

SKILL.md

Code Review

Perform a thorough senior-level code review on recent changes. Focus human review on complex logic, architecture, and edge cases. Delegate style enforcement to automated tools.

Related Skills: - kavak-documentation - Query for Kavak-specific patterns, logging/metrics standards, GitLab CI templates - Use kavak-platform/platform_docs_search MCP tool to verify Kavak best practices when reviewing

Quick Start

# 1. Get review range
BASE_SHA=$(git rev-parse HEAD~1)  # or: git merge-base main HEAD
HEAD_SHA=$(git rev-parse HEAD)

# 2. See what changed
git diff --stat $BASE_SHA..$HEAD_SHA
git diff --name-only $BASE_SHA..$HEAD_SHA --diff-filter=ACMR

# 3. After review, run quality gates (detect project type)

Quality gate commands by language:

LanguageLintBuildTest
Gogolangci-lint rungo build./...go test./...
Node/TSnpm run lintnpm run buildnpm test
Pythonruff check.python -m py_compilepytest
Java./mvnw checkstyle:check./mvnw compile./mvnw test

When to Review

This workflow applies to both self-reviews and peer reviews.

Mandatory: after completing a task, after implementing a feature, before merging.

Valuable: when stuck (fresh perspective), before refactoring (baseline), after fixing complex bugs.

Review Mindset

  • Thorough, not rushed - Read all related code before concluding
  • Evidence-based - Trace execution paths, don't assume bugs exist
  • Fix, don't just flag - Identify issues AND resolve them
  • Small scope - Review <400 lines at a time for effectiveness

Workflow

1. Scope the Diff

Identify all changed files:

git diff --name-only HEAD~1 --diff-filter=ACMR

For each file in the list, skip any that produces no actual diff hunks.

2. Understand Intent & Requirements

Before reviewing code, understand what it should do:

  • Original task/issue: What was requested?
  • Product impact: What does this deliver for users?
  • Acceptance criteria: What defines "done"?

Ask: "Does the implementation satisfy the requirements?"

3. Review Each File

For each changed file and each diff hunk, evaluate in context of the existing codebase.

MANDATORY: Before flagging Critical/Major, trace the complete execution path end-to-end. Confirm no defensive code prevents the issue. Consider error paths, cancellation, and edge cases.

4. Review Categories

Focus on: architecture (fits patterns, avoids coupling), correctness (edge cases, error handling), security (input validation, secrets, OWASP), performance (N+1 queries, memory), logging/metrics (cardinality explosions). Delegate style to linters.

Full checklist with all 11 categories: references/checklist.md Logging and metrics standards: references/logging-metrics-review.md

5. Check Project Rules

MANDATORY: Check and enforce rules from:

  • .claude/CLAUDE.md or CLAUDE.md (root)
  • .cursor/rules/ folder
  • AGENTS.md
  • Follow ALL patterns and conventions defined there

6. Report & Fix Issues

For each validated issue:

  • File: <path>:<line-range>
  • Issue: One-line summary
  • Fix: Concise suggested change

Severity levels:

  • Critical (P0): Security vulnerabilities, data loss, crashes
  • Major (P1): Significant bugs, performance issues, architectural violations
  • Minor (P2): Code style, minor improvements
  • Enhancement (P3): Nice-to-have improvements

MANDATORY: Fix all issues immediately after identifying them.

  1. Start with highest priority (Critical → Major → Minor)
  2. For each issue:

- Fix the code - Verify fix doesn't break anything - CHECKPOINT COMMIT: git add -A && git commit -m "fix: brief description"

  1. Continue until ALL issues are resolved

7. Run Quality Gates

After all issues are fixed, run lint → build → test for project type (see Quick Start table).

If any gate fails, fix immediately and commit the fix.

8. Final Report

Provide structured output:

### Strengths

[What's well done - be specific with file:line references]

### Issues Found & Fixed

- **Critical**: [count] - [brief list]
- **Major**: [count] - [brief list]
- **Minor**: [count] - [brief list]

### Quality Gates

- Lint: ✓/✗
- Typecheck: ✓/✗
- Build: ✓/✗
- Tests: ✓/✗

### Assessment

**Ready to proceed?** [Yes / Yes with notes / No - needs fixes]
**Reasoning:** [1-2 sentence technical assessment]

If no issues: "Code review complete. No issues found. All quality gates pass. Ready to proceed."

References

ReferencePurpose
references/checklist.mdDetailed review checklist
references/severity-guide.mdHow to classify issue severity
references/common-issues.mdCommon issues (TypeScript/Node)
references/common-issues-go.mdCommon issues (Go)
references/security-owasp.mdOWASP Top 10 security checklist
references/feedback-guide.mdHow to give constructive feedback
references/logging-metrics-review.mdLogging format and metrics cardinality standards
references/self-review-workflow.mdAutomated self-review during development

Best Practices

  1. No assumptions - Read ALL related code before flagging issues
  2. Fix immediately - Don't just report, fix and commit
  3. Checkpoint commits - Commit each fix separately for easy rollback
  4. Verify first - Trace execution paths before claiming bugs exist
  5. Project rules priority - Always check .cursor/rules/ and AGENTS.md
  6. Be timely - Review promptly to avoid blocking teammates
  7. Limit scope - Review <400 lines at a time for effectiveness

Principle: Review → Fix → Verify → Commit. A review that only identifies problems without fixing them is incomplete.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

04

需要参考平台分布和安装热度时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Claude Code

27.3%
按下载量换算83

trae

23.44%
按下载量换算71

Antigravity

17.16%
按下载量换算52

Codex

13.91%
按下载量换算42

windsurf

8.07%
按下载量换算25

Gemini CLI

3.53%
按下载量换算11

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

external-service

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

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。

来源信息

继续浏览同类 Skills