← All Tools
Prompt Injection Playground
A defensive reference for AppSec: how prompt-injection techniques work, and how to harden an LLM app against each one.
🔒 100% client-side — no model, nothing sent.
There is no LLM here and no network request. The evaluator below is a small set of naive keyword/pattern checks for teaching only — it is not a real model and not a production filter. “No match” does not mean an input is safe, and getting an input past these demo rules proves nothing about a real system.
How to use this page
- Browse the catalogue on the left: pick a technique family to see a worked example, why it works, concrete defenses and a reference.
- Read the defenses for each — this page is about hardening your app, not attacking others.
- Try the evaluator: paste text (or hit “Try this technique’s example”) to see which naive demo rules it trips. Remember the limits listed under it.
- Export the full catalogue (JSON/Markdown) for a wiki or ticket, or export your last evaluation.
Heuristic defense evaluator
See which bundled demo rules an input trips. This is keyword/pattern matching for education — not a real LLM and not a detection you should rely on.