Integrate Connector HTTP Use
Use this guide when wiring a trusted Session supervisor or SDK to an implemented connection relay. Follow the binding contract and wire profile; ordinary HTTPS_PROXY configuration alone does not implement credential injection.
- Resolve the agent's connection declarations at admission. Supply trusted binding selections for mode and explicit destination when needed. Refuse missing/ambiguous destinations; never infer an API host from token or discovery endpoints. Different origins need distinct declaration bindings and grants.
- Prepare and activate the attempt-bound grant with the immutable binding and attested confirmation public key. Keep the private signing capability in the trusted supervisor. Agent code receives a host-bound client interface, not a provider token or the signing key.
- Bind that client to the admitted declaration/destination. Reject caller Authorization, Cookie, relay-control, framing and transport-owned overrides. Apply the finite request limits and canonical path policy before signing.
- Construct the exact relay target and context, digest the complete bounded body and sign every forwarded end-to-end header under RFC9421. The proxy independently checks live policy, grant, logical upstream egress, destination and credential role; a signed caller request is not policy authority.
- Preserve provider HTTP status/body within the response policy and distinguish the relay-owned provenance/failure headers. Treat unsupported responses, timeouts and unknown outcomes as possible provider effects. Do not automatically mint a fresh invocation or replay a write after refresh.
- Run the checked-in conformance artifacts and then independent production adapter tests, including cross-replica replay, custody denial, restore, parser ambiguity, response leakage and live revocation. Do not claim transparent interception or runtime isolation from a schema example.
