Prism-Eval interactive demo

Catch non-deterministic LLM tool-call failures, prompt injections, and digit drops before production. Toggle a vulnerable vs hardened agent and run the G4 suite in your browser.

AI agent testing Digit drop detection Prompt injection red-team CI/CD gate

Results

Pass rate

—

Mean determinism

—

Critical false accepts

—

G4 invariant

—

Case	Attack	Status	Det	Reasons

Click “Run G4 adversarial suite” to execute the browser demo.

Local Python demo with the real engine: pip install "prism-eval[demo]" && streamlit run demo/app.py · InsightITS · GitHub · PyPI