Need safe upgrade/backport guidance for production affected by Codex compact 404 on 2026.5.12

We are an affected production deployment and need operational guidance after #123706 was closed as already implemented on current main.

Related issue: #123706

Production state

• OpenClaw: 2026.5.12 (f066dd2)
• Runtime/session model: openai/gpt-5.5 via openai-codex OAuth
• Fallback: openai/gpt-5.4-mini
• Explicit compaction override was configured:
  • agents.defaults.compaction.model = "google/gemini-3.1-pro-preview"
  • safeguard compaction enabled
  • truncateAfterCompaction=true
• User-facing failure, via Telegram:
  • Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

Observed behavior

Despite the explicit Gemini compaction model, failed turns attempted the Codex compact endpoint:

https://chatgpt.com/backend-api/codex/responses/compact

and failed with:

unexpected status 404 Not Found: {"detail":"Not Found"}
FailoverError
model_not_found

The fallback model then hit the same Codex compact endpoint and failed the same way. Gateway/Telegram health, VPS resources, and task queue were otherwise healthy.

What #123706 established

The review on #123706 says current main already fixes the relevant compaction owner-boundary behavior. It also says the inspected latest release, v2026.7.1-2, does not contain the relevant current-main fix.

Relevant commit mentioned in the review: ee6c2cb50df0.

Question

What is the safest recommended path for a production deployment currently on 2026.5.12?

1. Wait for the next release containing ee6c2cb50df0;
2. Upgrade to a current-main build;
3. Cherry-pick/backport the compaction routing fix;
4. Apply a config workaround to avoid this broken Codex compact route until a release ships.

We can keep using /new and reduced context as a temporary mitigation, but we would prefer not to move a production gateway to current main without maintainer guidance.

Current mitigation

We reduced automatic context injection and tool/skills payload sizes and restarted the gateway. A simple Telegram test passed after the mitigation, but the underlying release gap remains.

Start by reading related issue #123706 and its review, then inspect commit ee6c2cb50df0 against release v2026.7.1-2 and the 2026.5.12 deployment. Determine whether the safest supported recommendation is a release upgrade, current-main build, backport, or configuration workaround, and document the temporary mitigation and production risks.