[Feature] PROME Care Core structured-action guardrail integration

What Would You Like to See with the Gateway?

We would like to propose a PROME Care Core integration for checking proposed agent actions against fixed safety constraints, and confirm the appropriate contribution path before implementing a Portkey-specific connector.

Care Core is a hosted, key-protected API that accepts structured measurements supplied by the customer's application. It returns a selected action or a hold, with reasons. It does not generate text, classify arbitrary prompts, establish whether supplied measurements are true, grant permission, or execute actions. Its decision checker does not call a language model.

Product overview: https://prome.ai/care-core API: https://prome.ai/api/guardrail (customer key required)

Context for your Request

The intended integration would let an application check a proposed tool action before it is executed, using facts from trusted application systems rather than allowing an agent to certify its own risk. The application must still enforce current permissions and prevent the agent from bypassing the check.

We have a working standalone API, but have not yet released or verified a Portkey connector. We would like to confirm:

1. Is a custom webhook plus integration documentation the preferred route, or should we contribute a named guardrail integration?
2. Which supported hook is appropriate for checking a proposed tool action against trusted application-supplied measurements?
3. Can failOnError: true together with deny: true reliably stop the request for timeouts, non-200 responses, invalid JSON and invalid/missing verdicts? We would include these cases in connector tests before calling it supported.

Only the connector and documentation would be contributed. The proprietary core would remain hosted by PROME; no credentials or core source are included here.

This is an integration proposal, not a claim of an existing partnership or production-qualified Portkey integration.