Agent Governance Protocol
API ReferenceDecision APIEscalations

Get escalation notice

GET
/agp/escalations/{escalationId}

Authorization

AuthorizationBearer <token>

In: header

Scope: agp.decision.read, agp.decision.write

Path Parameters

escalationId*string

Response Body

application/json

application/json

application/json

application/json

curl -X GET "https://example.com/agp/escalations/string"
{  "type": "EscalationNotice",  "escalation_id": "string",  "task_id": "string",  "raising_agent_id": "string",  "raising_agent_role": "executive",  "concern_type": "ethical_violation",  "description": "string",  "evidence_refs": [    "string"  ],  "requested_action": "quarantine",  "resolution": {    "resolved_by": "string",    "outcome": "upheld",    "notes": "string",    "resolved_at": "2019-08-24T14:15:22Z"  },  "raised_at": "2019-08-24T14:15:22Z",  "signature": {    "alg": "string",    "kid": "string",    "issuer": "string",    "issued_at": "2019-08-24T14:15:22Z",    "payload_hash": "string",    "value": "string"  }}
{  "error": {    "code": "AGP_INVALID_REQUEST",    "message": "string",    "retryable": true,    "details": {}  }}
{  "error": {    "code": "AGP_INVALID_REQUEST",    "message": "string",    "retryable": true,    "details": {}  }}
{  "error": {    "code": "AGP_INVALID_REQUEST",    "message": "string",    "retryable": true,    "details": {}  }}

Raise escalation notice POST

Any authorized agent may raise an EscalationNotice (the Whistleblower mechanism). The raising agent does not have to be the executing agent. requested_action determines mandatory task state transition: - quarantine -> QUARANTINED (immediate, no human required) - pause_and_review -> APPROVAL_PENDING - escalate_to_human -> APPROVAL_PENDING with mandatory human approver EscalationNotices are permanent audit records even if later dismissed. See §23.

Resolve escalation notice PATCH

Populated by the resolving authority after review. Upheld, dismissed, or partially upheld.