fix(codex): keep large-request ttfb watchdog active

agent/chat_completion_helpers.py

@@ -534,25 +534,28 @@ def interruptible_api_call(agent, api_kwargs: dict):
             and _ttfb_disable_above > 0
             and _est_tokens_for_codex_watchdog >= _ttfb_disable_above
         ):
-            _ttfb_enabled = False
-            logger.info(
-                "Disabling openai-codex no-byte TTFB watchdog for large request "
-                "(context=~%s tokens >= %.0f). Waiting for backend response instead. "
-                "Set HERMES_CODEX_TTFB_STRICT=1 to force early reconnects.",
-                f"{_est_tokens_for_codex_watchdog:,}",
-                _ttfb_disable_above,
-            )
-        else:
-            _ttfb_cap = _env_float("HERMES_CODEX_TTFB_MAX_SECONDS", 120.0)
-            if _ttfb_cap > 0 and _ttfb_timeout > _ttfb_cap:
+            _large_request_ttfb_timeout = _codex_idle_timeout_default
+            if _ttfb_timeout < _large_request_ttfb_timeout:
                 logger.info(
-                    "Capping openai-codex no-byte TTFB timeout from %.0fs to %.0fs "
-                    "(context=~%s tokens). Set HERMES_CODEX_TTFB_MAX_SECONDS to tune.",
+                    "Scaling openai-codex no-byte TTFB watchdog from %.0fs to %.0fs "
+                    "for large request (context=~%s tokens >= %.0f). "
+                    "Set HERMES_CODEX_TTFB_STRICT=1 to keep the smaller cutoff.",
                     _ttfb_timeout,
-                    _ttfb_cap,
+                    _large_request_ttfb_timeout,
                     f"{_est_tokens_for_codex_watchdog:,}",
+                    _ttfb_disable_above,
                 )
-                _ttfb_timeout = _ttfb_cap
+                _ttfb_timeout = _large_request_ttfb_timeout
+        _ttfb_cap = _env_float("HERMES_CODEX_TTFB_MAX_SECONDS", 120.0)
+        if _ttfb_cap > 0 and _ttfb_timeout > _ttfb_cap:
+            logger.info(
+                "Capping openai-codex no-byte TTFB timeout from %.0fs to %.0fs "
+                "(context=~%s tokens). Set HERMES_CODEX_TTFB_MAX_SECONDS to tune.",
+                _ttfb_timeout,
+                _ttfb_cap,
+                f"{_est_tokens_for_codex_watchdog:,}",
+            )
+            _ttfb_timeout = _ttfb_cap
 
     _codex_idle_enabled = _codex_watchdog_enabled
     _codex_idle_timeout = _env_float(

tests/agent/test_codex_ttfb_watchdog.py

@@ -353,8 +353,8 @@ def test_ttfb_disabled_via_env_zero(tmp_path, monkeypatch):
 
 def test_large_codex_request_waits_instead_of_ttfb_reconnect(tmp_path, monkeypatch):
     """Large Codex inputs can legitimately take longer than the small-request
-    first-byte cutoff before the first SSE frame. Preserve the full input and
-    wait instead of killing/retrying at TTFB."""
+    first-byte cutoff before the first SSE frame. Scale the TTFB timeout up
+    for those requests instead of killing/retrying at the small-request cutoff."""
     from agent import chat_completion_helpers as h
 
     agent = _make_codex_agent(tmp_path, monkeypatch)
@@ -386,6 +386,45 @@ def test_large_codex_request_waits_instead_of_ttfb_reconnect(tmp_path, monkeypat
     assert "codex_ttfb_kill" not in closes
 
 
+def test_large_codex_request_can_still_ttfb_reconnect_when_capped(tmp_path, monkeypatch):
+    """Large Codex requests should keep a finite TTFB watchdog instead of
+    disabling it entirely. A low max cap should still force an early reconnect."""
+    from agent import chat_completion_helpers as h
+
+    agent = _make_codex_agent(tmp_path, monkeypatch)
+    monkeypatch.setenv("HERMES_CODEX_TTFB_TIMEOUT_SECONDS", "1")
+    monkeypatch.setenv("HERMES_CODEX_TTFB_MAX_SECONDS", "1")
+
+    closes: list = []
+    dummy_client = SimpleNamespace()
+    monkeypatch.setattr(agent, "_create_request_openai_client", lambda **k: dummy_client)
+    monkeypatch.setattr(
+        agent, "_abort_request_openai_client", lambda c, reason=None: closes.append(reason)
+    )
+    monkeypatch.setattr(
+        agent, "_close_request_openai_client", lambda c, reason=None: closes.append(reason)
+    )
+
+    stop = {"flag": False}
+
+    def fake_hang(api_kwargs, client=None, on_first_delta=None):
+        deadline = time.time() + 30
+        while time.time() < deadline and not stop["flag"] and not agent._interrupt_requested:
+            time.sleep(0.02)
+        raise RuntimeError("connection closed")
+
+    monkeypatch.setattr(agent, "_run_codex_stream", fake_hang)
+
+    large_input = "x" * 44_000  # ~11k estimated tokens, above the large-request gate.
+    try:
+        with pytest.raises(TimeoutError) as excinfo:
+            h.interruptible_api_call(agent, {"model": "gpt-5.5", "input": large_input})
+        assert "TTFB threshold: 1s" in str(excinfo.value)
+        assert "codex_ttfb_kill" in closes
+    finally:
+        stop["flag"] = True
+
+
 def test_large_codex_request_strict_ttfb_env_still_reconnects(tmp_path, monkeypatch):
     """Operators can force the old early-reconnect behavior for large inputs
     with HERMES_CODEX_TTFB_STRICT=1."""