fix(codex): size and propagate timeouts for Responses-API requests; lower stale defaults

Codex / Responses-API requests had three latent timeout bugs that combined
into the long silent hangs reported on #21444:

1. The non-stream stale-call detector estimated context tokens from
   ``api_kwargs["messages"]`` only. Codex / Responses-API payloads carry
   their conversational load in ``input`` (with ``instructions`` and
   ``tools``), so every Codex turn logged ``context=~0 tokens`` and the
   detector never applied its >50k / >100k tier bumps.

2. ``providers.<id>.request_timeout_seconds`` was silently dropped on the
   main Codex path. The chat_completions path and the auxiliary Codex
   adapter both forwarded it; the main path skipped it through three
   places (``build_api_kwargs``, ``ResponsesApiTransport.build_kwargs``,
   ``_preflight_codex_api_kwargs``).

3. The streaming stale detector had the same payload-shape bug for
   ``codex_responses`` requests, which route through the non-streaming
   detector (it's the path that emits the user-facing
   "No response from provider for 300s (non-streaming, ...)" warning that
   reporters keep pasting).

This commit:

- Adds ``estimate_request_context_tokens`` in ``chat_completion_helpers``,
  used by both the non-stream and stream detectors. Handles ``messages``
  (Chat Completions), ``input + instructions + tools`` (Responses API),
  bare lists, and an unknown-dict fallback.
- Forwards ``timeout`` through ``ResponsesApiTransport.build_kwargs``
  and ``_preflight_codex_api_kwargs`` (with guards against
  zero/negative/inf/bool values), and wires
  ``_resolved_api_call_timeout()`` into the Codex branch of
  ``build_api_kwargs``.
- Lowers the implicit non-stream stale defaults so fallback providers
  kick in faster when upstream stalls:
    * base   300s -> 90s
    * >50k   450s -> 150s
    * >100k  600s -> 240s
  These only apply when the user has *not* set
  ``providers.<id>.stale_timeout_seconds`` or
  ``HERMES_API_CALL_STALE_TIMEOUT``. Explicit config still wins.
- Adds regression tests for the estimator shapes, the new defaults, the
  context-tier scaling, transport timeout pass-through, and preflight
  timeout pass-through / rejection of invalid values.

Closes #21444
Supersedes #21652 #24126 #31855

Co-authored-by: Hoang V. Pham <26063003+hehehe0803@users.noreply.github.com>

tests/agent/test_non_stream_stale_timeout.py

@@ -0,0 +1,192 @@
+"""Tests for the non-stream stale-call detector context estimator.
+
+Covers:
+- ``estimate_request_context_tokens`` for Chat Completions, Responses API,
+  bare lists, and mixed-shape dicts.
+- ``AIAgent._compute_non_stream_stale_timeout`` with both legacy ``messages``
+  list and full ``api_kwargs`` dicts.
+- The May 2026 default-base change (300s -> 90s) and the lowered
+  context-tier ceilings (450/600 -> 150/240).
+"""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+import pytest
+
+
+def _write_config(tmp_path: Path, body: str) -> None:
+    hermes_home = tmp_path
+    (hermes_home / "config.yaml").write_text(body or "{}\n", encoding="utf-8")
+
+
+def _make_agent(tmp_path: Path, **overrides):
+    from run_agent import AIAgent
+    kwargs = dict(
+        model="gpt-5.5",
+        provider="openai-codex",
+        api_key="sk-dummy",
+        base_url="https://chatgpt.com/backend-api/codex",
+        quiet_mode=True,
+        skip_context_files=True,
+        skip_memory=True,
+        platform="cli",
+    )
+    kwargs.update(overrides)
+    return AIAgent(**kwargs)
+
+
+# ── estimator ──────────────────────────────────────────────────────────────
+
+
+def test_estimator_chat_completions_messages():
+    from agent.chat_completion_helpers import estimate_request_context_tokens
+    payload = {
+        "model": "gpt-5.4",
+        "messages": [
+            {"role": "user", "content": "x" * 400},
+            {"role": "assistant", "content": "y" * 400},
+        ],
+    }
+    # 800+ chars from messages -> ~200 tokens (char/4 estimate)
+    assert estimate_request_context_tokens(payload) >= 200
+
+
+def test_estimator_responses_api_input():
+    from agent.chat_completion_helpers import estimate_request_context_tokens
+    payload = {
+        "model": "gpt-5.5",
+        "instructions": "i" * 1000,
+        "input": "x" * 4000,
+        "tools": [{"name": "t", "description": "d" * 200}],
+    }
+    # input(4000) + instructions(1000) + tools (~stringified) -> well over 1000 tokens
+    tokens = estimate_request_context_tokens(payload)
+    assert tokens >= 1200, f"Responses API estimator returned {tokens}"
+
+
+def test_estimator_responses_api_long_session_triggers_tier():
+    """A real long Codex session (large ``input``) should clear the 50k boundary."""
+    from agent.chat_completion_helpers import estimate_request_context_tokens
+    payload = {
+        "model": "gpt-5.5",
+        "input": "x" * 240_000,  # ~60k tokens (240k chars / 4)
+        "instructions": "s" * 4000,
+    }
+    assert estimate_request_context_tokens(payload) > 50_000
+
+
+def test_estimator_bare_list_back_compat():
+    from agent.chat_completion_helpers import estimate_request_context_tokens
+    messages = [
+        {"role": "user", "content": "x" * 800},
+    ]
+    assert estimate_request_context_tokens(messages) >= 200
+
+
+def test_estimator_empty_inputs():
+    from agent.chat_completion_helpers import estimate_request_context_tokens
+    assert estimate_request_context_tokens({}) == 0
+    assert estimate_request_context_tokens([]) == 0
+    assert estimate_request_context_tokens(None) == 0
+
+
+def test_estimator_unknown_dict_fallback():
+    from agent.chat_completion_helpers import estimate_request_context_tokens
+    payload = {"random_field": "z" * 400}
+    assert estimate_request_context_tokens(payload) > 50
+
+
+# ── default base + tier scaling ────────────────────────────────────────────
+
+
+def test_default_base_is_90s(monkeypatch, tmp_path):
+    """Default base stale timeout dropped from 300s to 90s (May 2026)."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    (tmp_path / ".env").write_text("", encoding="utf-8")
+    monkeypatch.delenv("HERMES_API_CALL_STALE_TIMEOUT", raising=False)
+    _write_config(tmp_path, "")
+
+    agent = _make_agent(tmp_path)
+    base, implicit = agent._resolved_api_call_stale_timeout_base()
+    assert base == 90.0
+    assert implicit is True
+
+
+def test_short_codex_request_uses_base_only(monkeypatch, tmp_path):
+    """Codex payload below 50k tokens -> default 90s base."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    (tmp_path / ".env").write_text("", encoding="utf-8")
+    monkeypatch.delenv("HERMES_API_CALL_STALE_TIMEOUT", raising=False)
+    _write_config(tmp_path, "")
+
+    agent = _make_agent(tmp_path)
+    payload = {"model": "gpt-5.5", "input": "hi", "instructions": ""}
+    assert agent._compute_non_stream_stale_timeout(payload) == 90.0
+
+
+def test_long_codex_request_bumps_to_50k_tier(monkeypatch, tmp_path):
+    """Codex payload > 50k tokens -> at least 150s."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    (tmp_path / ".env").write_text("", encoding="utf-8")
+    monkeypatch.delenv("HERMES_API_CALL_STALE_TIMEOUT", raising=False)
+    _write_config(tmp_path, "")
+
+    agent = _make_agent(tmp_path)
+    payload = {"model": "gpt-5.5", "input": "x" * 240_000, "instructions": ""}
+    timeout = agent._compute_non_stream_stale_timeout(payload)
+    assert timeout >= 150.0
+    assert timeout < 240.0
+
+
+def test_very_long_codex_request_bumps_to_100k_tier(monkeypatch, tmp_path):
+    """Codex payload > 100k tokens -> at least 240s."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    (tmp_path / ".env").write_text("", encoding="utf-8")
+    monkeypatch.delenv("HERMES_API_CALL_STALE_TIMEOUT", raising=False)
+    _write_config(tmp_path, "")
+
+    agent = _make_agent(tmp_path)
+    payload = {"model": "gpt-5.5", "input": "x" * 500_000, "instructions": ""}
+    assert agent._compute_non_stream_stale_timeout(payload) >= 240.0
+
+
+def test_chat_completions_long_messages_bumps_tier(monkeypatch, tmp_path):
+    """Chat Completions estimator still works for the legacy messages path."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    (tmp_path / ".env").write_text("", encoding="utf-8")
+    monkeypatch.delenv("HERMES_API_CALL_STALE_TIMEOUT", raising=False)
+    _write_config(tmp_path, "")
+
+    agent = _make_agent(
+        tmp_path,
+        provider="openai",
+        base_url="https://api.openai.com/v1",
+        model="gpt-5.4",
+    )
+    payload = {
+        "model": "gpt-5.4",
+        "messages": [{"role": "user", "content": "x" * 240_000}],
+    }
+    assert agent._compute_non_stream_stale_timeout(payload) >= 150.0
+
+
+def test_explicit_user_config_overrides_default(monkeypatch, tmp_path):
+    """If the user explicitly sets a stale_timeout, the new defaults don't apply."""
+    monkeypatch.setenv("HERMES_HOME", str(tmp_path))
+    (tmp_path / ".env").write_text("", encoding="utf-8")
+    _write_config(tmp_path, """\
+providers:
+  openai-codex:
+    stale_timeout_seconds: 1800
+""")
+    monkeypatch.delenv("HERMES_API_CALL_STALE_TIMEOUT", raising=False)
+
+    import importlib
+    from hermes_cli import timeouts as to_mod
+    importlib.reload(to_mod)
+
+    agent = _make_agent(tmp_path)
+    assert agent._compute_non_stream_stale_timeout({"input": "hi"}) == 1800.0

tests/agent/transports/test_codex_transport.py

@@ -452,3 +452,64 @@ class TestCodexNormalizeResponse:
         tc = nr.tool_calls[0]
         assert tc.name == "terminal"
         assert '"command"' in tc.arguments
+
+
+
+class TestCodexTransportTimeout:
+    """Forward per-request timeout from build_kwargs to the SDK kwargs."""
+
+    def test_positive_timeout_preserved(self, transport):
+        kw = transport.build_kwargs(
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            tools=[],
+            timeout=600.0,
+        )
+        assert kw.get("timeout") == 600.0
+
+    def test_zero_timeout_dropped(self, transport):
+        kw = transport.build_kwargs(
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            tools=[],
+            timeout=0,
+        )
+        assert "timeout" not in kw
+
+    def test_none_timeout_omitted(self, transport):
+        kw = transport.build_kwargs(
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            tools=[],
+            timeout=None,
+        )
+        assert "timeout" not in kw
+
+    def test_inf_timeout_dropped(self, transport):
+        kw = transport.build_kwargs(
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            tools=[],
+            timeout=float("inf"),
+        )
+        assert "timeout" not in kw
+
+    def test_bool_timeout_dropped(self, transport):
+        """``True`` is technically int but must not survive — caller bug guard."""
+        kw = transport.build_kwargs(
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            tools=[],
+            timeout=True,
+        )
+        assert "timeout" not in kw
+
+    def test_request_overrides_can_supply_timeout(self, transport):
+        """request_overrides["timeout"] is honored when no explicit kwarg passed."""
+        kw = transport.build_kwargs(
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            tools=[],
+            request_overrides={"timeout": 450.0},
+        )
+        assert kw.get("timeout") == 450.0