unvibecode README

项目的 README,来自 github.com/FinanceFlash/unvibecode。合并为一个 markdown 文件:README.md。

UnvibeCode

Provide a brain for your codebase

Reverse engineer a complex codebase into business workflows, connected code, edge cases, and evidence-backed risks.

⭐ Hit Star to help increase UnvibeCode's visibility among developers.

Unvibe complex code. Trace what the business actually does

Quality checks PyPI version Python 3.11+ License GitHub stars

FinanceFlash/unvibecode | Trendshift

UnvibeCode AI codebase analysis demo showing connected code, business workflows, and risk findings

Your LLM helped write 100,000 lines of code. Now how do you understand what it actually built?

Reading files one by one does not show the complete business workflow. Pasting a large repository into an LLM can also lose the connections between entry points, state changes, dependencies, edge cases, and business outcomes.

UnvibeCode reverse-engineers a complex codebase into business workflows, connected code, edge cases, and evidence-backed risks.

Try UnvibeCode

Requires Python 3.11 or newer.

Windows

py --version
py -m pip install --upgrade unvibecode
py -m unvibecode --help
py -m unvibecode review --repository "D:\path\to\repository"

Example:

py -m unvibecode review --repository "D:\Projects\customer-support-agent"

macOS

python3 --version
python3 -m pip install --upgrade unvibecode
python3 -m unvibecode --help
python3 -m unvibecode review --repository "/path/to/repository"

Example:

python3 -m unvibecode review --repository "/Users/yourname/Projects/customer-support-agent"

Linux

python3 --version
python3 -m pip install --upgrade unvibecode
python3 -m unvibecode --help
python3 -m unvibecode review --repository "/path/to/repository"

Example:

python3 -m unvibecode review --repository "/home/yourname/projects/customer-support-agent"

No activation key. No customer OpenAI API key. The repository path is the only required input.

Explore sample analyses

smolagents · PocketFlow · mini-swe-agent · django-payments

Trusted by engineers @ Vertigo, Kovai.co, Ancient, Tingle Software, DVARA and Many More

UnvibeCode, in their words

“The business workflow getting lost across 500 files is the real issue. AI can keep generating code, but once nobody can trace why the product behaves the way it does, the speed advantage disappears fast.”

Kavitha Ranganathan — Senior PM, Better

“By tracing a customer recovery workflow from the screen to the code, I found that the system could mark a customer as having promised to pay based on the last digit of their ID, not an actual promise”

Fernanda Teixeira — QA Engineer, Vertigo

One repository review. Four connected outputs.

UnvibeCode helps developers understand what a codebase does, how its components connect, and what could break — without manually tracing every file.

1. Business Workflow Map — Understand what the system does

Reconstruct business workflows from source code. Explore execution paths, decisions, dependencies, state changes, and the implementation behind each workflow.

Useful for: onboarding, reverse engineering, understanding legacy applications, and exploring unfamiliar repositories.

Business Workflow Map

2. Connected Code Map — Trace where behavior lives

Explore interactive relationships between source files, imports, symbols, and dependencies. Select a file, inspect its connections, and download focused code context for further investigation.

Useful for: debugging, dependency tracing, architecture exploration, and preparing connected code for AI assistants.

Connected Code Map

3. Business Risk Findings — Discover what could break

Identify potential business-impacting implementation defects supported by source evidence. Explore affected workflows, implementation behavior, potential impact, remediation, and verification checks.

Useful for: investigating business logic defects, reviewing AI-generated applications, and identifying fragile workflows.

Not every repository produces confirmed risk findings. UnvibeCode distinguishes supported findings from cases where evidence is insufficient.

Business Risk Findings

4. Complete Repository Context — Take the analysis further

Download a normalized repository ZIP for code review, documentation, or further AI-assisted investigation. Need less code? Use the Connected Code Map to retrieve a focused context package around a selected file.

Useful for: engineering handoffs, repository exploration, and reusable source context.


Why not just dump your entire codebase into an LLM?

More source code doesn't automatically mean better code understanding.

Before: full-code dumps into an LLM. After: structured code relationships, graph-aware context, and evidence-gated findings with UnvibeCode.


How does UnvibeCode compare?

Tool Primary strength UnvibeCode difference
Aider AI-assisted code editing Repository-wide workflow and risk investigation without an editing task
Repomix Repository packaging for AI Interactive code maps, business workflows, and evidence-backed risks
Qodo PR-Agent Pull-request review Understand existing application behavior beyond a proposed code change
CodeQL Static security and correctness analysis Business workflow explanations connected to implementation and operational consequences

When should you use UnvibeCode?

Understanding an unfamiliar repository

Discover business workflows, connected components, and implementation paths without manually opening every file.

Reviewing AI-generated applications

Understand what was actually implemented and investigate potential business logic defects.

Debugging across multiple files

Trace dependencies and surrounding code before changing a function in isolation.

Investigating application risks

Review source-backed findings and understand potential operational consequences.

Preparing context for AI coding assistants

Download complete repository context or focused connected code for further investigation.


Supported source languages

Connected-code mapping and context preparation recognize:

Language File types
Python .py
JavaScript .js, .jsx, .mjs, .cjs
TypeScript .ts, .tsx
Rust .rs
PHP .php
Ruby .rb
HTML and CSS .html, .htm, .css
Go .go

Analysis depth varies by language and repository structure. See limitations for details, and Go parsing and connected graphs for how Go packages, files, and call edges are resolved.


Privacy

UnvibeCode uses lightweight safeguards to reduce accidental credential sharing and keep AI requests focused on supported analysis tasks. Before the official UnvibeCode client sends code for AI analysis, it looks for recognizable credentials—such as API keys, passwords, access tokens and private keys—and replaces detected values with placeholders. Deeply encoded or custom secret format files could be skipped by the safeguards

Documentation

Contributing

Contributions are welcome in documentation, testing, examples, and supported tooling.

Support and feedback

Found a problem or have a feature request? Open a GitHub issue.

For product questions, repository reviews, or collaboration, contact [email protected].

Security Policy

Report a vulnerability

If you discover a security vulnerability in UnvibeCode, please report it privately using the link below.

Report a vulnerability privately

Include the affected version, steps to reproduce, and potential impact. Please do not disclose security vulnerabilities in public GitHub issues.

License

UnvibeCode uses a mixed licensing model. Materials published under Apache License 2.0 retain that license, including commercial-use and redistribution rights. The official Python package may also contain separately licensed, proprietary business-domain definitions, workflow archetypes, and analysis assets.

Commercial use of the official package is permitted for internal business activities, subject to the restrictions on extraction and separate redistribution of proprietary assets.

Read the Licensing Clarification for the specific scope, permissions, restrictions, and package-metadata explanation.


Contributors & Acknowledgements

A special thanks to the developers who have helped strengthen UnvibeCode:

We appreciate their thoughtful engineering contributions and their commitment to making UnvibeCode more reliable for developers.

unvibecode (FinanceFlash/unvibecode)

正在搜索数据库…

这是什么?UnvibeCode 会将复杂的代码库逆向还原为业务工作流、关联代码图谱、边界情况以及有证据支撑的风险发现,帮助开发者在不必逐个阅读文件的情况下,了解大型或 AI 生成的代码库究竟在做什么。
它会像 Aider 那样修改我的代码,还是只做分析?它是一个只读的调查工具,而不是编辑器。与 Aider 不同,它不会修改代码,而是还原工作流、梳理依赖关系,并在你现有的仓库中找出风险。
我该如何运行它?安装官方客户端(Python 3.11+),并将其指向你的仓库路径,这是唯一必需的输入。README 分别给出了 Windows、macOS 和 Linux 的运行示例。
我需要自己的 OpenAI API 密钥或激活码吗?不需要。项目明确说明,你既不需要激活密钥,也不需要用户自备的 OpenAI API 密钥,只需 Python 3.11 或更高版本。
它对所有编程语言的分析效果都一样好吗?关联代码映射目前仅识别 Python、JavaScript、TypeScript、Rust、PHP、Ruby、HTML/CSS 和 Go。分析的深度取决于具体语言以及仓库的结构。
我的源代码在被分析时会发生什么?在将代码发送给 AI 之前,客户端会扫描可识别的凭据(如 API 密钥、密码和私钥),并将其替换为占位符。但它也指出,经过深度编码或采用自定义格式的密钥可能会漏网。
仓库审查会生成哪四种输出?一份描述系统功能的业务工作流图、一份展示文件与符号之间关系的关联代码图、一份有源代码证据支撑的业务风险发现,以及一个用于后续 AI 辅助工作的完整仓库上下文 ZIP 包。
它和 Repomix 或 CodeQL 有什么不同?Repomix 只是将仓库打包以供 AI 使用,CodeQL 则进行静态安全分析。而 UnvibeCode 会生成可交互的代码图谱、还原的业务工作流,以及与实际运营后果相关联的、有证据支撑的风险发现。
在自己运行之前,我能先看看输出是什么样子的吗?可以。README 中链接了几个真实开源仓库的示例分析:smolagents、PocketFlow、mini-swe-agent 和 django-payments。
大小
12.2 MB
星标
1.9K
时长
3个月
提交 · 90 天