PR #17544: Fix Codex auxiliary fallback model drift (gpt-5.2-codex → gpt-5.4)

• Repository: NousResearch/hermes-agent
• Author: afurm
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/17544

Description (problem / solution / changelog)

What does this PR do?

This PR updates Hermes-Agent’s Codex auxiliary fallback model to a currently supported Codex model so ChatGPT-account users can continue to use auxiliary features (compression, vision extraction, etc.) when payment/credit fallbacks are triggered.

Problem resolved:

• _CODEX_AUX_MODEL was still pinned to gpt-5.2-codex, which is rejected by ChatGPT-account Codex backends.
• This caused auxiliary calls routed through _try_payment_fallback to fail with HTTP 400 “model is not supported” and pause retry logic in long sessions.

This update also removes brittle test coupling by asserting against _CODEX_AUX_MODEL instead of hardcoded strings.

Related Issue

Fixes #17533

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• agent/auxiliary_client.py
  • Update _CODEX_AUX_MODEL from "gpt-5.2-codex" to "gpt-5.4".
  • Refresh nearby inline comments describing current Codex fallback model behavior.
  • Keep selection logic intact (_try_codex and auxiliary provider chain unchanged).
• tests/agent/test_auxiliary_client.py
  • Import _CODEX_AUX_MODEL in tests.
  • Replace hardcoded "gpt-5.2-codex" assertions with _CODEX_AUX_MODEL for resilient assertions.

How to Test

1. Run focused tests:
   • scripts/run_tests.sh tests/agent/test_auxiliary_client.py -k codex
2. Add temporary logging/trace in a ChatGPT-account Codex environment and trigger an auxiliary fallback path (for example: force payment-fallback path and call compression/vision auxiliary task).
3. Confirm no "model is not supported" errors for Codex auxiliary calls and that fallback returns a successful response.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: <!-- e.g. Ubuntu 24.04, macOS 15.2, Windows 11 -->

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

• Please include auxiliary fallback logs showing successful Codex responses with gpt-5.4 if CI/manual verification is run in a ChatGPT-account Codex environment.

Changed files

• agent/auxiliary_client.py (modified, +5/-6)
• tests/agent/test_auxiliary_client.py (modified, +12/-11)

PR #17765: fix(aux): remove hardcoded Codex fallback model, drop Codex from auto chain

• Repository: NousResearch/hermes-agent
• Author: teknium1
• State: open | merged: False
• Link: https://github.com/NousResearch/hermes-agent/pull/17765

Description (problem / solution / changelog)

Summary

Deletes _CODEX_AUX_MODEL and removes _try_codex from the auxiliary fallback chain. Codex-OAuth is no longer a second-order fallback that guesses a model ID — if the user's main provider fails, the chain tries OpenRouter / Nous / custom / api-key and stops, instead of rolling through a hardcoded Codex model that OpenAI silently rotates out of its ChatGPT-account allow-list.

Closes #17533. Supersedes #17544 (which reset the stale-constant clock rather than removing it).

Why not the one-line constant bump (#17544)

That constant has already drifted twice in 6 weeks:

ChatGPT-account Codex allow-list is undocumented and OpenAI publishes no changelog. Any pinned default rots — the question is just "how soon."

Changes

Behavior

Validation

tests/agent/                                 2220 passed
tests/run_agent/test_provider_parity.py       113 passed
tests/run_agent/test_provider_fallback.py       9 passed
tests/run_agent/test_compressor_fallback_update.py   passed
E2E: removed symbols gone                    PASS
E2E: chain is 4 rungs, no openai-codex       PASS
E2E: payment fallback returns (None, None)   PASS
E2E: Step 1 Codex-main user path unaffected  PASS (gpt-5.4, gpt-5.5 both honored)
E2E: explicit model=None warns, returns None PASS

Credit: @pokibao (#17533) for the test matrix + root-cause analysis, @afurm (#17544) for the constant-bump proposal that surfaced this as worth a proper fix.

Changed files

• agent/auxiliary_client.py (modified, +65/-26)
• tests/agent/test_auxiliary_client.py (modified, +39/-25)
• tests/agent/test_codex_cloudflare_headers.py (modified, +7/-6)
• tests/run_agent/test_provider_parity.py (modified, +12/-4)