# How do you guys security test your own app with claude code without it thinking ur a hacker? — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/a9765023334f7987

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-10-26 04:20 ingested 2026-10-27 02:36

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1xfn487/how_do_you_guys_security_test_your_own_app_with/)

## retained text



```
Solo dev here, building a self hosted app where AI agents do work for you. so security is kind of a big deal for me lol

Before I release it I want to actually try and break it myself. things like auth bypass, one user getting into another users data, privilege escalation, and prompt injection against the agents.

Claude code is really good at reviewing code but when I ask it to actually *try* the attack, like write the payload or a script to exploit something it found, sometimes it gets a bit cautious. which I get, it doesnt know its my app. but its slowing me down

What do you guys do?

* do you just tell it upfront its your own code running locally? does putting that in CLAUDE.md help at all
* do you ask for failing tests instead ("write a test that proves user A cant read user B's stuff")? is that just as good
* or do you run proper tools  and just get claude to go through the results
* anyone found a good way to test prompt injection on there own agents?

Not trying to hack anyone else, I just want it solid before other people start running it. what's worked for you.

Also maybe I'm afraid of getting banned also... that's why I'm a little hesitant asking it to do stuff like that. Or worse flagged by the authority.
```



document fetched 2026-10-27 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1xfn487/how_do_you_guys_security_test_your_own_app_with
