Noir Guardrails Playground

A client-side mock-proxy playground to author rails/config and see how the guardrails trigger — no API keys required.

[01] / Rail Authoring

Rail Logic
Saved: (none)

[02] / Prompt Evaluation

User Prompt
Mode:
Editor Mode:
Schema Snippets

[03] / Results

Output Idle
00 Privacy Injection SQL Schema Abuse Policy
Radar idle — run a prompt to profile risk.
(results will appear here)

[04] / Trace

Trace Inspector Idle
  • (trace will appear here)

[04B] / Bifrost Topology

Network Topology
Switch to Bifrost mode to visualize request routing.
Validated your policy? Move to Safety as Code to generate a GitHub Action for this rail. Generate CI/CD Gate

[05] / Validation

Schema Validator (AJV)
If your rail logic contains a JSON Schema under "schema", the mock output will be validated and trace will show validation results.

[06] / Red Team Challenge

Can You Beat the Rails?
Goal: craft a malicious prompt that still gets a Pass (Success) through the high-security rails. Defensive goal: keep attacks blocked and earn hall-of-fame entries.
Challenge idle. Load the rails to begin.
Attempts0
Bypasses0
Blocked0
Hall of Fame — Blocked Payloads
Most creative malicious prompts your rails successfully stopped.
  1. No blocked payloads yet — go break things (safely).
Wall of Deflected Attacks
Live feed of elite blocks — high-signal attacks pinned with the exact rail logic that shut them down.
  1. Feed is quiet. Land a high-value deflection to claim first post.