Aurelius: The AI Operator

Overview

Aurelius is the AI operator persona for the Praetorian Guard Platform, named after the Stoic philosopher-emperor Marcus Aurelius. When Agent Mode is enabled, Aurelius serves as the top-level orchestrator that commands a team of specialized offensive security agents to conduct reconnaissance, scanning, and attack operations across your attack surface.

Aurelius assesses your security posture, plans attack paths, delegates to specialist agents, and coordinates multi-step offensive campaigns. It operates within the same rules of engagement as human Praetorian operators: all targets must exist in the platform, all operations require confirmation, and no denial-of-service or availability attacks are permitted.

How Aurelius Works

When you switch the AI Assistant to Agent Mode (indicated by a red badge), Aurelius follows a structured operational workflow:

1. Assess — Query the attack surface to understand scope and current security posture
2. Analyze — Identify highest-value targets, weakest links, and viable attack paths
3. Plan — Propose an attack plan with specific capabilities, targets, and expected outcomes
4. Confirm — Present the plan to the operator for approval before executing
5. Execute — Run approved operations and report findings with evidence of impact
6. Report — Deliver results focused on material risk demonstrated and remediation guidance

Specialist Agent Team

Aurelius delegates specialized tasks to purpose-built agents using the spawn_agent mechanism. Each agent is an autonomous operator with its own tools, skills, and operational constraints:

Agent Dispatch Mechanism

Aurelius uses two dispatch paths depending on the execution context:

• Synchronous dispatch — For compute-context operations, sub-agents execute in-process and return results directly to Aurelius. This enables real-time multi-step workflows where one agent's output feeds the next.
• Asynchronous dispatch — For gateway-context operations, sub-agents are queued via SQS for background processing. Results are persisted to the conversation and available on the next polling cycle.

Agent nesting is limited to 5 levels deep to prevent runaway recursion. Each agent has a configurable maximum of 25 tool calls per execution to bound operational scope.

Direct Tools

In addition to delegating to specialist agents, Aurelius can directly use:

Rules of Engagement

• No denial-of-service — Cannot degrade or disrupt client systems
• Strict scope enforcement — Only operates against assets present in the platform
• No production modification — Cannot alter configurations, delete data, or change system state
• Operator confirmation required — Always confirms before executing active scans or offensive capabilities
• Tool execution limits — Maximum 3 retries per tool, 10 tools before progress summary

Prerequisites

• AI Inference must be enabled
• AI Agents must be enabled (for non-Praetorian users)
• User must hold Analyst or Admin role

Accessing Agent Mode

1. Open the AI Assistant
2. Click the mode toggle to switch from Query (blue) to Agent (red)
3. Begin your conversation with an operational objective
4. Aurelius will assess, plan, and request confirmation before executing any active operations