openclaw - 💡(How to fix) Fix [Bug] openclaw doctor --fix still rewrites openai-codex/* → openai/* on v2026.5.22, breaking Codex-OAuth-only compaction (re-occurrence of #78407)

Summary

This is a re-occurrence of #78407 and #79461, confirmed still present on v2026.5.22.

After running openclaw doctor --fix (or after an update-triggered doctor --non-interactive --fix), all openai-codex/* model references are rewritten to openai/* with agentRuntime.id: "codex" added. For users authenticated only via OpenAI Codex OAuth (no OPENAI_API_KEY), this breaks compaction entirely:

1. Agent chat still works — Codex OAuth intercepts those calls correctly
2. Compaction fails — Codex native compaction cannot bind a thread, falls back to "context engine", which tries openai/gpt-5.5 directly → "No API key found for provider openai"
3. The agentRuntime.id: "codex" workaround does not fix compaction because the Codex harness logs: "ignoring OpenClaw compaction overrides for Codex app-server compaction" then "codex app-server compaction could not use thread binding" → falls back to context engine → same failure

Environment

• openclaw v2026.5.22 (npm global install, upgraded from v2026.5.4)
• Ubuntu 26.04 LTS
• Auth profiles: openai-codex:<chatgpt-account> + ollama:default. No openai:* profile.
• No OPENAI_API_KEY

Reproduction

1. Configure working openclaw.json with:

"model": { "primary": "openai-codex/gpt-5.5" }

1. Run openclaw doctor --fix
2. Doctor rewrites to:

"model": { "primary": "openai/gpt-5.5" },
"models": { "openai/gpt-5.5": { "agentRuntime": { "id": "codex" } } }

1. Compaction triggers → fails with:

[agent/embedded] ignoring OpenClaw compaction overrides for Codex app-server compaction
[agent/embedded] codex app-server compaction could not use thread binding
[agent/embedded] native harness compaction could not use its session binding; falling back to context engine
[compaction-diag] provider=openai/gpt-5.5 outcome=failed detail=No_API_key_found_for_provider_openai

1. Even agents.defaults.compaction.provider: "ollama" is ignored — Codex harness intercepts compaction first.

Manual Workaround

Restore openai-codex/ prefix and remove agentRuntime:

"model": { "primary": "openai-codex/gpt-5.5", "fallbacks": ["ollama/glm-5.1"] },
"models": { "openai-codex/gpt-5.5": {} }

Impact

• openclaw doctor --fix is destructive for Codex-OAuth-only users on v2026.5.22
• Every doctor run requires manual config repair
• The agentRuntime.id: "codex" workaround only fixes agent chat — it does not fix compaction
• Setting compaction.provider: "ollama" as fallback is ignored because Codex harness takes priority

Related

• #78407 — original report (closed)
• #79461 — v2026.5.7 re-report (closed/locked)
• #78491 — Codex runtime model refs break auth profile selection
• #78519 — silent billing switch from ChatGPT to OpenAI API