Open-source Apex security firewall between AI agents and Salesforce data. Wraps any @InvocableMethod with five-gate enforcement.

Five-gate chain: SchemaValidator → AccessGate → RateLimiter → RollbackGuard → AuditPublisher. No bypass path, fail-closed by design.

Real running-user CRUD/FLS enforcement via WITH USER_MODE — agents cannot read/mutate fields the user is not permissioned for.

Declarative policy config in Guard_Policy__mdt: schema contracts (JSON), rate budgets, profile/perm-set scoping, record ceilings.

Transactional safety: Savepoint boundary + per-call record ceiling stops bulk-injection attacks from landing partial state.

Real-time observability: Lightning dashboard (guardAuditDashboard) + sf agentguard CLI plugin watching AgentGuard_Audit__e stream.

117/117 tests passing (83% packaged coverage). STRIDE-mapped threat model with six threat categories tied to specific gates.

Zero dependencies, MIT licensed. Built for Agentforce, MCP-connected assistants, and custom LLM integrations calling Apex.