Configuration

All detection patterns and thresholds are in resk2/config/patterns.yaml.

Sections

• global: languages, confidence, max input
• direct_injection: high/medium/low patterns
• bypass_detection: jailbreak + stealth patterns
• memory_poisoning: memory manipulation + fake facts
• goal_hijack: drift, scope, escalation
• exfiltration: endpoints, data collection, encoding, webhook
• inter_agent_injection: masquerade, role override, chain, trust
• vector_similarity: backend, threshold, attack_patterns
• acl_decision_tree: RBAC policy tree
• content_framing: syntactic masking, sentiment, oversight, persona
• thresholds: per-detector scoring parameters