Hatena
Curated: If You're Worried About Security, Copy This Question into Claude Code (20)
SoftwareFirst seen 167 points
A Japanese tech article shares a long list of prompts you can paste into a coding AI agent, such as Claude Code, to check a project for security problems. The prompts cover login, permissions, input handling, abuse, and data exposure.
What the article shares
The author gives a set of ready-made questions. You copy them into a coding AI agent and point it at a code repository. The agent then reads the code and looks for common security weak spots. The prompts are grouped into six areas: login, permissions, injection, abuse, exposure, and a final review step.
Key terms in plain words
A coding AI agent is a tool that can read and edit code on its own. A repository is the folder that holds a project's code. The prompts ask the agent to check how users sign in, what they are allowed to do, and what data is visible to the public. The article also asks the agent to review its own findings, so that it does not report false alarms as real bugs.
Why it matters
Many security bugs come from simple mistakes, such as trusting a user ID sent by the browser or leaving a test page open. A checklist like this can help developers find these mistakes early. However, an AI agent can also miss problems or report wrong ones. Developers should still test their findings and confirm them before fixing anything.
What to watch
- Whether the prompts find real bugs in real projects
- How well AI agents avoid false alarms when checking code
- Whether teams add these checks to their normal review process
- Whether the article's advice to confirm each issue before filing it is followed
Glossary
- Claude Code
- An AI coding tool that can read, write, and check code in a project.