fix(codex): drop dead model slugs that HTTP 400 on ChatGPT Pro

[teknium1]

Summary

Remove three dead Codex-CLI slugs from the openai-codex OAuth fallback list. The chatgpt.com Codex backend rejects them with HTTP 400 for every ChatGPT Pro account we tested live today.

Affected slugs:

• gpt-5.2-codex
• gpt-5.1-codex-max
• gpt-5.1-codex-mini

Backend response on every call:

HTTP 400 {'detail': "The '<slug>' model is not supported when using Codex with a ChatGPT account."}

Reproducible live against https://chatgpt.com/backend-api/codex/models?client_version=1.0.0 — the live list returns 6 slugs (gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.3-codex, gpt-5.3-codex-spark, gpt-5.2), none of which are the three removed here.

Why this matters

Whenever _fetch_models_from_api() falls back to DEFAULT_CODEX_MODELS (transient network failure, first-run before cache, no token yet) the picker surfaced these dead slugs and HTTP 400'd on selection. The forward-compat synthesis table chained them downstream too.

Changes

• hermes_cli/codex_models.py:
  • drop the 3 dead slugs from DEFAULT_CODEX_MODELS
  • drop them from _FORWARD_COMPAT_TEMPLATE_MODELS driver/synthetic positions
  • explanatory comment so they don't get re-added from the public Codex CLI catalog
• tests/hermes_cli/test_codex_models.py: pivot the forward-compat test driver from gpt-5.2-codex (now removed) to gpt-5.3-codex (which is templated by 4 entries — gpt-5.5, gpt-5.4-mini, gpt-5.4, gpt-5.3-codex-spark — so still exercises the synthesis path)

The Copilot, opencode-zen, openai (public API) and openai-api lists in hermes_cli/models.py still reference these slugs — those providers may accept them, scope is intentionally limited to openai-codex OAuth.

Validation

• Live verified chatgpt.com/backend-api/codex/models returns 6 slugs, none matching the 3 removed
• Targeted suite: 285/285 pass across test_codex_models.py, test_cli_provider_resolution.py, test_fast_command.py, test_model_metadata.py, test_model_validation.py, test_copilot_auth.py, test_image_generation_plugin_dispatch.py
• E2E: get_codex_model_ids() with no API access now returns ['gpt-5.5', 'gpt-5.4-mini', 'gpt-5.4', 'gpt-5.3-codex', 'gpt-5.3-codex-spark'] — all 5 live-verified working against the OAuth backend earlier today
• Live test E (separate session): 5/5 working models pass identical prompts in 1.84-5.03s; same test confirmed the 3 removed slugs all crash with HTTP 400

Forward path

If OpenAI re-enables these on the Codex backend, _fetch_models_from_api() will pick them up automatically — no code change needed. The defaults list is only consulted when live discovery is unavailable.

[github-actions]

🔎 Lint report: hermes/hermes-5bf34d29 vs origin/main

ruff

Total: 0 on HEAD, 0 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 0 pre-existing issues carried over.

ty (type checker)

Total: 9513 on HEAD, 9513 on base (➖ 0)

🆕 New issues: none

✅ Fixed issues: none

Unchanged: 5010 pre-existing issues carried over.

Diagnostics are surfaced as warnings — this check never fails the build.