Credit Analysis & Risk Agent - Behavioral Trust Monitor
Normal: Clean credit analysis run with no policy violations
V1 - Unauthorized Tool: Agent attempts to call a blocked tool (retail_customer_ledger_query)
V2 - Rate Limit: Agent exceeds API rate limits (61 calls vs 50 limit)
V3 - Spend Gate: Agent recommends £75k facility, triggering human approval gate
V4 - MCP Rug-pull: Agent detects compromised MCP tool with hash mismatch
V5 - Intent Drift: Agent performs out-of-scope subsidiary queries
Full Sequence: Runs all 5 violation scenarios in order with 8-second delays