[AgentShield] Add a feature for risk score tracking

We've been building [AgentShield](https://useagentshield.com) — an observability platform focused on AI agent safety rather than just tracing. After talking to teams running agents in production, we noticed everyone monitors what happened after a failure. Nobody predicts what's about to go wrong. So we built three features around that gap: --- ### 🔮 Risk Score (0-1000) A continuously updated score per agent based on: - Alert rate (30d) - Hallucination frequency - Error rate - Cost stability - Approval compliance Think of it as a **credit score for your AI agent**. 800+ = reliable. Below 200 = shouldn't be in production. --- ### 💰 Pre-Execution Cost Prediction Before your agent runs a task, we estimate cost based on historical patterns (p25, p50, p95). If your support bot usually costs $0.40-$1.20 per interaction but suddenly the prediction shows $4.80, something changed. You catch it **before** burning budget. --- ### 💥 Blast Radius Calculator Estimates the **maximum potential damage** an agent can cause based on: - Permissions and tool access - Action history (destructive vs read-only) - Financial exposure (max transaction × daily volume) - Approval coverage gaps A read-only chatbot → blast radius near zero. An agent with refund access processing $5K/day? That number matters. --- All three work across **LangChain, CrewAI, OpenAI Agents SDK**, and any framework via REST API or MCP integration. Free tier available. Curious what you all think — are these the right signals to track for production agents, or are we missing something?