Integration Pattern

SMERC adds recoverability scoring to OpenAI agent approvals

OpenAI Agents approval hooks can pause or reject sensitive tool calls. SMERC adds a pre-execution decision: whether the proposed action is recoverable enough to execute, should proceed with constraints, needs human review, should freeze, or should be denied.

ALLOW THROTTLE FREEZE DENY ESCALATE
Adapter folder Adapter code Example report

What It Shows

A real approval callback can use SMERC as the decision brain

The reference adapter converts an agent tool call into a SMERC runtime permission request, evaluates recoverability, and returns an approval action for the host application: approve, approve with constraints, route to human review, or reject.

Tool call SMERC request Posture Approval action Replay packet

Output Mapping

SMERC posture becomes approval behavior

ALLOWApprove and log replay.
THROTTLEApprove with constraints such as preview, scope limit, rate limit, or recovery path.
ESCALATERequire explicit human review before execution.
FREEZE / DENYReject or pause until more context exists.

Boundary

Reference adapter, not production authorization

This adapter is designed to show the integration pattern. Production use would need identity, customer policy, durable state, audit retention, secrets handling, and security review. It is not an OpenAI-endorsed integration.

SMERC vs AI guardrails AI deployment governance MCP scanner