Summary

During setup, ChatGPT OAuth completed and provider config was set to openai-codex, but Codex CLI was not installed automatically. Runtime then failed until Codex CLI was installed manually.

I also switched Hermes terminal backend between local and docker during setup/testing. That may have contributed, but setup/installer should still handle this transition gracefully and ensure Codex
dependency/auth state is valid afterward.

Environment

• Hermes: v1.0.0
• OS: Ubuntu 24.04.4 LTS
• Initial state: codex binary absent from PATH
• Backend changed during setup/testing: local <-> docker

Repro

1. Start on a system without Codex CLI installed.
2. Configure Hermes for openai-codex and complete ChatGPT OAuth.
3. Switch backend (local/docker) during setup/testing.
4. Start Hermes usage requiring Codex path.
5. Observe runtime failures until Codex CLI is manually installed.
6. Install Codex CLI manually (npm install -g @openai/codex) and retry -> succeeds.

Expected behavior

Setup should robustly validate dependencies/auth across backend transitions:

• Ensure Codex CLI is installed (or auto-install it), and
• Ensure OAuth/provider state remains valid after backend change, or
• Fail fast with a clear remediation message.

Actual behavior

OAuth/config appears successful, but dependency gap is discovered later at runtime.

Request

Please make openai-codex setup resilient to backend switching (local/docker) and enforce Codex CLI dependency checks/install during setup (not deferred to runtime).