Docs / Integrations / Semantic Kernel
Apply gVner AI checks in Semantic Kernel filters.
Attach gVner AI to SK function pipelines so high-risk functions require explicit ALLOW decisions.
Where to hook intent checks
Invoke intent check in function filter middleware before function execution.
Always call before side effects, not after.
Pass correlation IDs for traceability.
Payload mapping to /adapters/http/intent
{
"tenant_id": "acme",
"runtime_id": "semantic-kernel-01",
"agent_id": "agent-01",
"intent": "update_customer",
"requested_scopes": ["update_customers"],
"tool": "crm",
"context": {"source": "semantic-kernel", "workflow": "demo"},
"requested_budget": 1.0,
"work_type": "execution",
"correlation_id": "req-123"
}
Fail-closed behavior
Required: if gVner AI is unreachable, returns DENIED, or response is invalid, do not execute. Treat as blocked and log audit context.
Sample wrapper pattern
# in SK filter
decision = gvner_evaluate(intent)
if decision["decision"] != "ALLOWED":
raise PermissionError(decision["reason"])
# continue to target function
Relevant console pages
Evidence and exports
Export evidence packets from Exports.
Use Receipts for integrity checks.
Use Regulator Packets for external review.