fix(codex): make the reasoning summary level configurable#33222

Problem

T3 asks Codex for detailed reasoning summaries on every turn. Some OpenAI-compatible API gateways accept only auto and reject the request, so no turn can run through them in T3.

Change

Codex instances get a reasoning summary setting under the runtime section (detailed, auto, concise or none). The adapter sends the selected value as the turn's summary.

For managed ChatGPT instances, the runtime section showed only its own read-only path fields, so the new setting would have been missing there. It now shows below them. Binary path, CODEX_HOME path, shadow home path and launch arguments stay hidden in managed mode, as before.

T3 already requests reasoning summaries on every turn, and the setting only picks their level. It defaults to detailed, so existing setups behave the same unless someone changes it.

Verification

• vp test run apps/server/src/orchestration-v2/Adapters/CodexAdapterV2.test.ts: the existing turn-params table checks that each value is sent and that a missing setting sends detailed.
• vp run check:changed: format, lint, knip, typecheck and the affected tests pass.
• Manual, against an OpenAI-compatible gateway through a Codex instance using an existing CLI: detailed is rejected, auto works.

UI Change

On main:

With the change:

Managed ChatGPT instance (not signed in; layout only):

Written with GPT-6.1 Sol in Codex and reviewed with Claude Opus 5.5 in Claude Code, both running in T3 Code.