Test AI agents in the browser you already use¶
Stealth Prompt is a browser-native assistant for authorized AI security testing. Pick one chat interaction, generate an evidence-driven test message, inspect the reply, and continue the same line of investigation without moving cookies or recreating the UI in an automation script.
Install from Chrome Web Store Run the guided demo
Authorized targets only
Use Stealth Prompt only on systems you own or are explicitly authorized to assess.
Why it exists¶
-
Real browser context
Test the authenticated UI and account state already open in Chrome.
-
Exact interaction scope
Detect or select the input, send control, and response container. The model cannot invent a browser operation or silently broaden the target.
-
Progressive autonomy
Generate only, approve each send, prepare guided follow-ups, or authorize a bounded automatic run.
-
Evidence-backed results
Model judgement can raise a potential finding. Confirmation requires a deterministic check or an explicit operator decision.
Two ways to connect AI¶
| Path | Best for | Providers | Evidence |
|---|---|---|---|
| Local Core | CLI access, credential isolation, deterministic scorers, durable artifacts | Claude CLI, Codex CLI, Ollama, OpenAI API, Fake | JSON and self-contained HTML on disk |
| Direct API | Fastest setup with no local service | OpenAI API, Anthropic API | Up to 50 browser-local reports and JSON download |
The assessment loop¶
Observe → hypothesize → generate → review → send → capture → verify → report
Stealth Prompt is deliberately closer to a focused Repeater for an AI-agent interface than to a broad automatic scanner. The operator chooses the page, interaction, objective, data-sharing policy, and autonomy limits.