Internal Components
ShadowAudit is organized around a small runtime gate and supporting infrastructure for audit, replay, integrations, and reporting.
Agent Framework Adapter
│
▼
Core Gate Engine
│
├─ Capability Mapper
├─ Policy Engine
├─ Risk Evaluation
├─ Enforcement FSM
└─ Audit Logger
Core gate
The gate is the runtime enforcement primitive. It receives an agent ID, capability, task context, risk category, and payload, then returns a deterministic decision.
Policy engine
The policy engine evaluates explicit policy-as-code rules such as allow, deny, and require_approval.
Audit logger
The audit logger records decisions in SQLite with hash chaining and optional Ed25519 signatures.
Adapters
Adapters let ShadowAudit sit in front of common agent tool systems:
- LangChain
- LangGraph
- CrewAI
- OpenAI Agents SDK
- MCP
- Direct Python tools
Assessment and reporting
The scanner, simulator, and reporting modules support CI checks, replay, compliance reports, and governance evidence generation.