Prompt Injection Check.

Inspect pasted text or a supported file for hidden content and instruction signals. Your file stays in this browser; the analysis does not send it anywhere.

LOCAL ANALYSISNO UPLOAD
WORKSPACE Choose an input below

Scan files

Up to 12 files at once · processed locally in your browser
Files to check

Limits

Limits: DOCX/PDF/PPTX/XLSX 10 MiB; EML 2 MiB; text files 128 KiB. PDF 200 pages / 2 million extracted characters. No OCR or complete visual hiding check.

Type or paste text

Type or paste · Ctrl + Enter to analyze

Type plain text, or paste from another app. When your browser supplies HTML with a paste, we also inspect its hidden textual layers. Editing afterward clears that HTML snapshot.

Plain text only; no HTML layer captured.

HOW IT WORKS4 steps · local browser analysis · exportable results
01
Select input

Paste text or choose one or more supported files from your device.

02
Analyze locally

Rule-based checks run in your browser. No account or server-side scanning of your content.

03
Review findings

Check the exact location and decoded excerpt before using the content with AI.

04
Export results

Save the selected result as JSON, TXT or Markdown, print it as PDF, or export the whole batch as a ZIP.

What the levels mean
HIGH
Hidden structural content or a strong instruction signal was found. Review the excerpt and its source location.Context helps distinguish an instruction from an example or a document feature.
INFO
An unusual element was found, such as an invisible character or a contextual phrase. Review its location.
NONE
No known patterns were found using the current analysis methods.Check the coverage summary for layers this scan could not inspect.
Limitations
  • Pasted plain text cannot reveal formatting in the original file. Scan the file itself when available.
  • Static, rule-based check. No AI model is used.
  • Hidden formatting and invisible characters are checked without relying on the text language. Instruction phrases in other languages may be missed.
  • No image analysis or OCR. PDF text, annotations, form fields and metadata are checked; attachments are inventoried without reading their contents. Complete visual hiding and embedded Office objects remain outside scope.
  • Some structural signals can be HIGH without an instruction phrase. Other document areas usually need an instruction match. Interpret each finding in its source context.
  • Findings can overlap. Base64 decoding reveals text for rule checks; review any resulting match in context.
  • Test version: results and rules may change.
Why I built this

I built this tool because I could not find one that did exactly what I needed: scan text and documents for prompt injection signals directly in the browser, without uploading private files to a server.

It started as a personal need and became my first project of this kind. I want to keep improving detection, file support, usability and testing while keeping the core tool free to use.

If Prompt Injection Check saves you time or helps you inspect a suspicious document before using it with AI, your support helps keep the project going.

Prompt Injection Check is an independent project maintained by AdrianRepair. Public methodology, limitations and release history document how the product works and changes over time.

Support & supporters

The core scanner is free to use, and scanned content is analyzed locally in your browser.

If Prompt Injection Check helps you, consider supporting its continued development.

☕ Buy me a coffee on PayPal

What does your support help with?

It contributes to hosting and domain costs, testing across browsers and file formats, maintaining local processing, improving detection and usability, and keeping the core tool free to use.