Summary

• preserve base_url, api_key, and api_mode when _spawn_background_review() forks a review agent
• prevent named custom providers resolved to provider='custom' from degrading into a bare custom/main review clone
• add regression coverage for memory, skill, and combined background-review modes

Problem

Gateway/main sessions using named custom providers are runtime-resolved to provider='custom' plus explicit runtime credentials. Background review cloned only model + provider, so the fork lost endpoint state and failed with:

• resolve_provider_client: custom/main requested but no endpoint credentials found
• Auxiliary background review failed: No LLM provider configured

Test Plan

• pytest tests/run_agent/test_background_review_runtime_inheritance.py tests/run_agent/test_background_review_summary.py -q
• minimal runtime repro: clone with only model + provider fails; clone with inherited runtime fields succeeds