Co-Pilot / 辅助式
更新于 25 days ago

candid

Rron-myers
0.0k
ron-myers/candid
86
Agent 评分

💡 摘要

Candid 是一个基于 Radical Candor 框架的 AI 驱动代码审查工具,提供可配置的反馈。

🎯 适合人群

寻求代码质量改进的软件开发人员希望强制执行编码标准的团队领导专注于生产准备的 DevOps 工程师寻找可操作见解的质量保证专业人员教授编码最佳实践的教育工作者

🤖 AI 吐槽:看起来很能打,但别让配置把人劝退。

安全分析低风险

风险:Low。建议检查:是否执行 shell/命令行指令;是否发起外网请求(SSRF/数据外发)。以最小权限运行,并在生产环境启用前审计代码与依赖。

Candid

A Claude Code plugin for configurable code reviews that combine thoroughness with actionable feedback. Based on Kim Scott's Radical Candor framework: Care Personally + Challenge Directly.

Overview

Candid provides AI-powered code reviews that catch issues before production while teaching you better patterns. Choose between Harsh (brutal honesty) or Constructive (caring + challenging) tone. Define project-specific standards in Technical.md that get enforced automatically. Every issue comes with concrete fixes rated by confidence level.

Core Workflow

The review process follows these steps:

  1. Run /candid-review on your changes
  2. Select your preferred tone (Harsh or Constructive)
  3. Candid reviews with full architectural context
  4. Get categorized issues with actionable fixes
  5. Select which issues to track as todos
  6. Apply fixes and optionally auto-commit

Key Features

  • Configurable Tone - Harsh (brutal honesty) or Constructive (caring + challenging)
  • Technical.md Support - Define and enforce project-specific coding standards
  • Focus Modes - Target reviews on security, performance, architecture, or edge cases
  • Actionable Fixes - Every issue includes concrete code with confidence ratings
  • Re-Review - Track progress on fixes across review sessions
  • Auto-Commit - Automatically commit applied fixes with detailed messages
  • Todo Integration - Convert issues to tracked todos with multi-select
  • Issue Categorization - Organized by severity (Critical → Architectural)

Installation

npx skills add https://github.com/ron-myers/candid

Then restart Claude Code.

Quick Start

Basic review:

/candid-review

With tone preset:

/candid-review --harsh
/candid-review --constructive

Focus on specific aspects:

/candid-review --focus security
/candid-review --focus performance

Auto-commit applied fixes:

/candid-review --auto-commit

Documentation

Full documentation at www.candid.tools:

Technical.md

Define project-specific standards that Candid enforces during reviews. Violations appear as 📜 Standards Violation in your review.

Quick setup:

/candid-init                    # Auto-generate from codebase (thorough analysis)
/candid-init react              # React-specific standards
/candid-init minimal            # Minimal starter
/candid-init --effort quick     # Fast analysis (~30 sec)
/candid-init --effort thorough  # Deep analysis (~3-5 min, default)

Or copy a template:

cp templates/Technical-minimal.md ./Technical.md cp templates/Technical-react.md ./Technical.md cp templates/Technical-node.md ./Technical.md

Keep it focused: under 500 lines, verifiable rules only. Skip what your linter handles. See the Custom Standards guide for detailed guidance.

Configuration

Persist your tone preference and other settings in config files:

User-wide default:

mkdir -p ~/.candid echo '{"tone": "harsh"}' > ~/.candid/config.json

Project-specific:

mkdir -p .candid echo '{"tone": "constructive", "autoCommit": true}' > .candid/config.json

See the Config Reference for all options including exclude patterns, focus areas, and mergeTargetBranches.

Philosophy

This plugin is built on the Radical Candor principle that the best feedback:

  1. Cares Personally - Shows understanding of context and difficulty
  2. Challenges Directly - Doesn't hedge or soften real issues

Whether you choose Harsh or Constructive tone, every review aims to find real issues before production, provide immediately applicable fixes, help you track what needs doing, and teach patterns that prevent future problems.

Updating

claude plugin update candid@candid

Then restart Claude Code. See CHANGELOG.md for version history.

Links

License

MIT License - see LICENSE for details.

Author

Ron Myers

五维分析
清晰度9/10
创新性8/10
实用性9/10
完整性9/10
可维护性8/10
优缺点分析

优点

  • 可配置的语气以实现个性化反馈
  • 自动执行编码标准
  • 提供带有信心评级的可操作修复
  • 针对性审查的焦点模式

缺点

  • 可能需要为 Technical.md 进行初始设置
  • 严厉的反馈可能不适合所有团队
  • 新用户的学习曲线
  • 依赖 Claude Code 环境

相关技能

pytorch

S
toolCode Lib / 代码库
92/ 100

“它是深度学习的瑞士军刀,但祝你好运能从47种安装方法里找到那个不会搞崩你系统的那一个。”

agno

S
toolCode Lib / 代码库
90/ 100

“它承诺成为智能体领域的Kubernetes,但得看开发者有没有耐心学习又一个编排层。”

nuxt-skills

S
toolCo-Pilot / 辅助式
90/ 100

“这本质上是一份组织良好的小抄,能把你的 AI 助手变成一只 Nuxt 框架的复读机。”

免责声明:本内容来源于 GitHub 开源项目,仅供展示和评分分析使用。

版权归原作者所有 ron-myers.