Anomaly Detection Safety Net (P5)
"A guardian angel for your AI."
The Anomaly Detection system (P5) is a built-in safety mechanism that protects your project from "Agent Loops" and accidental resource exhaustion. It sits between the AI Agent and your tools, monitoring every action in real-time.
Capabilities
1. Infinite Loop Prevention
If an Agent tries to call the exact same tool with the exact same arguments repeatedly (e.g., read_file("main.py") 50 times in a row), the Safety Net intervenes.
- Threshold: 50 consecutive identical calls.
- Action: Blocks the 51st call.
- Feedback: Returns ⛔ ANOMALY DETECTED: Loop detected to the Agent, forcing it to change strategy.
2. Smart Batching
The system is intelligent enough to distinguish between a Loop and a Batch Operation. - Scenario A: Reading the same file 50 times -> BLOCKED 🛑 - Scenario B: Reading 50 different files -> ALLOWED ✅
Configuration
You can adjust the sensitivity of the Safety Net via environment variables:
Monitoring
When an anomaly is triggered, it is recorded in the Usage Dashboard (boring-monitor) under the "Anomalies" tab (coming soon).