Dashboard GuidePrompt Shield

Prompt Shield

The Prompt Shield panel lets you manually test the 155-rule shield engine and manage the rule whitelist.

Live Input Scanner

Paste any text and click Analyze to see what the shield detects. Results show the verdict (ALLOW/REVIEW/BLOCK), score, and every rule that triggered with severity and confidence. Use the Load Demo Payload button to see what a real attack looks like.

Dashboard scans always run all 155 rules regardless of whitelist settings, so you can test detections that would normally be suppressed for internal traffic.

Rule Whitelist

Click Manage to expand the full list of 155 rules. Toggle rules on/off to whitelist them for internal traffic. Whitelisted rules are skipped when scanning agent system prompts (which legitimately reference files like SOUL.md).

Pre-configured whitelist:

Rule IDReason
COG-SOULAgent system prompts reference SOUL.md
COG-IDENTITYAgent system prompts reference IDENTITY.md
COG-MEMORYAgent system prompts reference MEMORY.md
COG-RULESAgent system prompts reference RULES.md
COG-TOOLS-MDAgent system prompts reference TOOLS.md
COG-AGENTS-MDAgent system prompts reference AGENTS.md
COG-OPENCLAW-JSONAgent configs reference openclaw.json
COG-GATEWAY-JSONAgent configs reference gateway.json
FIN-SWIFT-CODESWIFT/BIC regex matches common all-caps words

Recent Shield Events

The panel also displays recent shield scan results with verdict, score, and detection details for quick review.