研讨会演示了如何利用双会话 Claude Code 工作流精准定位代码漏洞并自动生成合并请求。

测试调试📅 2026/03/26
#Bug#开发者#GitHub#手动触发#中风险#可复用#半自动#代码仓库#报告#测试
开发者使用 Claude Code 为 Percolator 风险引擎生成定制审计提示词并自动创建漏洞修复合并请求
Ran our first OpenClaw workshop yesterday. 

11 people came to learn how to use AI to find bugs and contribute to Percolator.

Main takeaways from the call:

You do not need to be technical. 

Download Claude Code, clone a Percolator repo, and ask it what the most vulnerable parts of the system are -> Then ask it to write you a specific prompt for finding bugs in those areas. -> Paste that prompt in a new session. -> Claude reads the code, finds issues, writes the report, and can even open the PR for you.

The key insight @dcc_crypto shared: use one Claude session to craft the prompt, a separate session to actually do the audit. 

The first session gives you something specific what types of attacks to look for in the risk engine, the pricing engine  rather than a generic scan.

For running it 24/7, you want OpenClaw on a VPS. 

For one time audits and contributions, Claude Code is easier and has fewer security risks.

Percolator is open source. All the repos are public. The prompts and context are already in the code. 

You show up, clone the repo, and start asking questions.

Workshop recording will be shared for anyone who missed it.

Come find bugs.

https://t.co/q3BiqYtH9G