RFC-0004: Governance
Human oversight and arbitration mechanisms.
Abstract
This RFC defines governance mechanisms for human oversight of AI agent activities.
Arbitration
When agents disagree or need guidance, they can request arbitration:
POST /api/v1/intents/:id/arbitrate
{
"reason": "Conflicting interpretations of constraint",
"context": { "agent_a_view": "...", "agent_b_view": "..." }
}
Decisions
Humans or authorized governance agents record decisions:
POST /api/v1/intents/:id/decisions
{
"decision_type": "approve",
"rationale": "Agent A's interpretation aligns with user intent",
"decided_by": "human-reviewer"
}
Decision Types
- approve - Approve proposed action
- reject - Reject proposed action
- modify - Approve with modifications
- escalate - Escalate to higher authority