Govern · Human decisions
Make high-stakes approvals part of the record.
A manager approving a discount, an adjuster authorizing a payout, a clinician signing off — these are consequential actions too. CHP treats a human approval, consent, or denial as the same governed, provable event as an agent's tool call.
How it would work
Approval and denial, as first-class outcomes.
A capability that requires approval emits a structured outcome when consent is withheld — denial code, reason, and the evidence ID, tied to the same correlation as everything else in the session. The record shows not just what happened, but who was allowed to make it happen.
Demonstrated, not yet shipped: the approval and denial primitives exist in the protocol today. The human-in-the-loop workflow and UI are what we build with a design partner.
{
"invocation_id": "inv_session_abc_001",
"capability_id": "schedule_technician",
"capability_version": "1.0.0",
"correlation": { "correlation_id": "session-abc" },
"outcome": "denied",
"success": false,
"data": null,
"error": null,
"denial": {
"code": "entitlement_denied",
"message": "service:dispatch is required",
"retryable": false
},
"evidence_ids": ["evt_8f3a1c"],
"started_at": null,
"completed_at": "2026-06-16T15:14:22.104Z"
}Have a high-stakes approval flow?
Bring the decision that has to be defensible later. We will show you how it maps onto the protocol and build the human-in-command workflow with you.