fix(auxiliary): omit temperature for GPT-5 Responses

[AXEG0]

What does this PR do?

Makes auxiliary-request temperature handling route-aware for GPT-5 models.

When the resolved auxiliary route uses the Responses API (codex_responses), the request omits temperature, which that route rejects for affected GPT-5 models. Chat-completions routes—including custom providers and OpenRouter—continue to receive the caller's configured temperature. This keeps the fix aligned with the actual transport contract instead of applying a model-name-only rule.

Related Issue

N/A — no public issue is linked.

Related/overlapping work:

• fix(auxiliary): OMIT temperature for gpt-5.5 family (#51083) #51157 also changes GPT-5 temperature handling, but applies a provider-agnostic model predicate. This PR instead gates omission on the resolved API mode so chat-completions routes keep their supported temperature value.
• fix(flush_memories): strip temperature from codex_responses fallback #15620 and fix(auxiliary): universal retry when any provider rejects temperature #15627 are earlier fallback-oriented fixes for unsupported temperature parameters; this change prevents the invalid field on the verified Responses route before a request is sent.

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: carry the resolved API mode through sync, async, and retry request builders; omit GPT-5 temperature only for codex_responses; preserve existing Kimi and Arcee handling.
• tests/agent/test_auxiliary_client.py: add route-specific builder and request coverage proving Responses requests omit temperature while chat-completions requests preserve it.

How to Test

1. Run scripts/run_tests.sh tests/agent/test_auxiliary_client.py -q.
2. Verify GPT-5 auxiliary requests resolved to codex_responses omit temperature.
3. Verify the same model on chat_completions preserves the configured temperature.

Focused result: 178 passed.

Additional checks: ruff check ., git diff --check, and scripts/check-windows-footguns.py --all pass.

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 (focused affected suite passes; the full GitHub Actions matrix is awaiting maintainer approval)
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: Linux

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — N/A; no user-facing interface or behavior documentation changed
• I've updated cli-config.yaml.example if I added/changed config keys — N/A; no configuration keys changed
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A; no architecture or workflow changed
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — platform-independent Python change; Windows footgun check passes
• I've updated tool descriptions/schemas if I changed tool behavior — N/A; no tool schema changed

Screenshots / Logs

N/A — regression coverage and command results are listed above.

AI Assistance

OpenAI — GPT-5 via Codex CLI (implementation assistance, test execution, and PR-description drafting).

[teknium1]

Thanks for the focused regression coverage. Two issues need resolution before this can be salvaged.

Problems

• The added predicate matches gpt-5.4 (agent/auxiliary_client.py:408), but the unchanged test still expects that model to preserve temperature (tests/agent/test_auxiliary_client.py:3118-3132). The PR's test file therefore has conflicting expectations.
• The predicate has no provider, endpoint, or API-mode check. Main explicitly keeps GPT-5 on chat completions for Nous and generic custom providers (run_agent.py:1429-1437), and this helper also feeds the main chat path (agent/chat_completion_helpers.py:946-952). Meanwhile the Codex auxiliary Responses adapter already omits temperature (agent/auxiliary_client.py:971-972). The change needs to distinguish the affected LiteLLM/Responses route from non-Responses GPT-5 routes.

Suggested changes

• Reconcile the existing gpt-5.4 expectation with the intended contract.
• Make the omission route-aware and add request-level coverage for both the affected route and a GPT-5 chat-completions route.

Automated hermes-sweeper review.

[AXEG0]

Addressed both review points in fee1f188a2.

• agent/auxiliary_client.py:404-413: GPT-5 temperature omission is now gated on api_mode == "codex_responses"; chat-completions routes (Nous/custom) retain it. The resolved mode is threaded through sync, async, and retry request construction (:3437-3448, :6544-6548, :7147-7151). Codex adapter behavior remains unchanged (it independently omits temperature; :920-922).
• tests/agent/test_auxiliary_client.py:3172-3235: reconciled GPT-5.4 expectations and added request-level assertions for Responses omission and chat-completions preservation, including Nous/custom routes.

Validation: pytest -q tests/agent/test_auxiliary_client.py — 309 passed (37.98s); targeted route tests — 12 passed; ruff check passed.