openclaw - 💡(How to fix) Fix doctor: model ref warning should check auth profile before suggesting openai-codex/* → openai/* rewrite [2 comments, 3 participants]

Problem

openclaw doctor always prints:

openai-codex/gpt-5.4 should become openai/gpt-5.4
openai-codex/gpt-5.4-mini should become openai/gpt-5.4-mini

regardless of the user authentication method. Users who use OpenAI Codex OAuth (mode=oauth in auth.profiles) intentionally use the openai-codex/ prefix because their models are routed through the Codex app-server harness, not the PI OpenAI path.

The doctor --fix mode already checks whether Codex is installed and has usable OAuth before rewriting refs (it selects agentRuntime.id: "codex" when applicable). But the read-only warning emitted during openclaw doctor (without --fix) does not perform this pre-check. This creates confusion:

1. Users see a scary warning and think their config is broken.
2. Users who run doctor --fix expecting a fix may get an unnecessary migration on a working setup.
3. New OAuth users get misled about whether their config is correct.

Expected behavior

The read-only warning should check the user auth profiles before emitting:

• If auth.profiles contains an openai-codex entry with mode=oauth and the Codex harness is available → suppress the "should become" warning (the config is correct as-is), or emit a more nuanced message like: "openai-codex/ refs are valid for your Codex OAuth setup. Run doctor --fix to migrate to canonical openai/* refs with codex runtime if desired."*
• If no Codex OAuth is configured → keep the current warning (legacy refs that should be migrated).

Suggested approach

In the Codex route repair check doctor warning pass, read auth.profiles and check for a usable openai-codex OAuth profile before emitting the "should become" suggestion. The same auth-profile awareness that --fix already has should be applied to the read-only warning path.

Why it matters

New OpenClaw users setting up Codex OAuth get an unnecessary false alarm on their first openclaw doctor run. Experienced users who rely on Codex OAuth keep seeing warning noise on every health check, eroding trust in doctor diagnostic output.