Why Praetor

Why Praetor?

Modern software engineering moves at breakneck speeds, but auditing codebases for production readiness, operations, security, and compliance remains a slow, siloed, and error-prone process.

Traditional approaches fall into two categories, both of which have severe limitations:

1. Manual Compliance and Audit Checklists:
   • Problem: Takes weeks of developer and auditor time. Checklists become stale instantly. No direct coupling exists between the audit spreadsheet and the actual source code.
2. Generic Single-Agent LLM Code Reviews:
   • Problem: A single generic prompt lacks deep domain expertise across diverse disciplines (like WCAG accessibility, Prometheus metrics, and SOC2 access controls simultaneously). More importantly, single-agent models suffer from hallucinations, referencing functions, files, or vulnerabilities that do not exist.

---

The Praetor Solution: Specialized Swarm & Gated QC

Praetor solves these challenges through a three-pronged architecture:

1. The Expert Swarm (18 Personas)

Instead of relying on a single generalist model, Praetor splits the audit workload among 18 specialized autonomous personas. Each agent runs with a custom profile:

• Years of Experience: 10 to 20 years of simulated senior/staff expertise.
• Defined Mandate: Specific target areas (e.g., the Unit Test Agent only focuses on boundary conditions and mocked endpoints, while the Chaos Agent evaluates circuit breakers and timeouts).
• Tooling Adaptation: Automatic detection and syntax generation for your specific stack (e.g., emitting Datadog alerts if you use Datadog, Prometheus rules if you use Prometheus).

2. The 4-Judge Quality Council (QC)

All generated artifacts are submitted to the Quality Council before emission. If any artifact fails the quality gates, it is sent back to the originating agent with review feedback for regeneration.

• Judge 1 (Coverage): Validates that all critical paths identified in technical discovery are mapped.
• Judge 2 (Citations): Independently verify that every referenced file and line number exists and corresponds to the finding.
• Judge 3 (Clarity): Evaluates readability, structure, and actionability of the output.
• Judge 4 (Skip-Validity): Reviews any items deferred by agents to ensure they are genuinely out-of-scope or missing code.

3. Absolute Traceability

Every finding, risk, or test specification maps directly to source-code lines:

Finding: Access control check is missing on endpoint `/api/v1/payments/refund`
Status: READY
Attribution: A06 (Security Agent)
Citation: prompt/04-mandates/SECRET_SCAN_MANDATE.md:L142-L148

This guarantees an audit evidence pack that human auditors, compliance officers, and SREs can immediately verify and trust.