← sources

I made a free, browser-only checker for hidden text and injection in prompts you copy from the web

source: reddit (reddit:PromptEngineering)published 2026-11-14 04:08ingested 2026-11-15 02:36

retained text

I often copy prompts from blogs, READMEs and posts and paste them into Claude Code. Then I learned that text can be hidden with Unicode tag characters: the AI reads it, but you can't see it. So I made a small page that checks a prompt before I hand it over.

What it does:

\- Decodes hidden text (tag characters, zero-width spaces) and shows what it says

\- Looks for injection patterns in English and Japanese ("ignore previous instructions", reading \~/.ssh, sending data to outside URLs, \\x/\\u escapes, %XX, rot13)

\- No length limit, so text hidden at the very end is checked too

\- Everything runs in your browser. Nothing is sent.

It is pattern matching, so it will miss things. The page lists every rule it checks.

[https://ikasumiworks.com/en/prompt-check/](https://ikasumiworks.com/en/prompt-check/)

I'm a hobby developer in Japan and built it with help from Claude Code. If you find a prompt it misses, I'd like to hear about it.

**Edit (Nov 15):** v1.5.0 adds Chinese, Korean and Spanish checks, and now flags zero-width characters packed with hidden text.  
v1.5.1: when it finds warning signs, it no longer suggests which AI to use.

document fetched 2026-11-15 00:34 · canonical https://reddit.com/r/PromptEngineering/comments/1xvsu2v/i_made_a_free_browseronly_checker_for_hidden_text