Agent SafetyOutlineDraft

Anomalous Tool Use Detection: The Agent Firewall

Baseline normal tool frequency, then auto-block and page when a read-only agent suddenly looks like an admin.

Series draft — Part 9 of 15 in Hardened Agentic Stack. Outline only; expand before un-drafting.

Phase 3: Telemetry & Observability

The Problem

Agents find legitimate-but-unexpected paths that still violate policy.

The Infrastructure Fix

Usage baselining in Grafana/Loki; trip wires on volume and privilege jumps.

The Architecture Pattern

Statistical Behavioral Analysis — security as deviation from the norm.

Planned sections

  1. The “Oh No” moment — concrete incident or near-miss that makes the risk visceral.
  2. ClawQL context — how this control protects a high-privilege local/edge agent.
  3. Technical how-to — concrete configs, policies, or snippets a builder can apply.
  4. Safety check — what “trusted enough” looks like once this layer is in place.

Key visuals

  • 24h baseline vs spike alert

Source modules (docs.clawql.com)

Rule of Three (keep on publish)

LayerTakeaway
ProblemAgents find legitimate-but-unexpected paths that still violate policy.
Infrastructure fixUsage baselining in Grafana/Loki; trip wires on volume and privilege jumps.
Architecture patternStatistical Behavioral Analysis — security as deviation from the norm.

About the author

Daniel Smith builds ClawQL, an agent operating system for token-efficient discovery and execution over APIs — with observability, hardened tool boundaries, and production routing for LLM workloads. He writes here about the systems problems behind shipping agents.