diff --git a/agent/chat_completion_helpers.py b/agent/chat_completion_helpers.py
index 56228ac09241..8392a76a3f43 100644
--- a/agent/chat_completion_helpers.py
+++ b/agent/chat_completion_helpers.py
@@ -1241,7 +1241,17 @@ def try_activate_fallback(agent, reason: "FailoverReason | None" = None) -> bool
         _fb_is_azure = agent._is_azure_openai_url(fb_base_url)
         if fb_provider == "openai-codex":
             fb_api_mode = "codex_responses"
-        elif fb_provider == "anthropic" or fb_base_url.rstrip("/").lower().endswith("/anthropic"):
+        elif (
+            fb_provider == "anthropic"
+            or fb_base_url.rstrip("/").lower().endswith("/anthropic")
+            or base_url_hostname(fb_base_url) == "api.anthropic.com"
+        ):
+            # Custom providers (e.g. cron-anthropic) point at the native
+            # api.anthropic.com host with no "/anthropic" path suffix, so the
+            # name/suffix checks above miss them and they default to
+            # chat_completions → POST /v1/chat/completions → 404. Match the
+            # host the same way determine_api_mode() and _detect_api_mode_for_url()
+            # do on the primary path. (#32243, #49247)
             fb_api_mode = "anthropic_messages"
         elif _fb_is_azure:
             # Azure OpenAI serves gpt-5.x on /chat/completions — does NOT
diff --git a/hermes_cli/runtime_provider.py b/hermes_cli/runtime_provider.py
index 7f4692b838cd..690c2c961024 100644
--- a/hermes_cli/runtime_provider.py
+++ b/hermes_cli/runtime_provider.py
@@ -20,6 +20,7 @@
     DEFAULT_XAI_OAUTH_BASE_URL,
     PROVIDER_REGISTRY,
     _agent_key_is_usable,
+    _nous_inference_env_override,
     format_auth_error,
     resolve_provider,
     resolve_nous_runtime_credentials,
@@ -92,6 +93,13 @@ def _detect_api_mode_for_url(base_url: str) -> Optional[str]:
 
     - Direct api.openai.com endpoints need the Responses API for GPT-5.x
       tool calls with reasoning (chat/completions returns 400).
+    - Direct api.anthropic.com endpoints must use the native Messages
+      API (``/v1/messages``).  Anthropic also exposes an OpenAI-compat
+      ``/chat/completions`` shim on the same host, but Pro/Max OAuth
+      subscriptions are only billed against the native Messages route;
+      hitting the shim accounts against a separate "extra usage" pool
+      that is empty by default and surfaces as HTTP 400 "You're out of
+      extra usage."  See issue #32243.
     - Third-party Anthropic-compatible gateways (MiniMax, Zhipu GLM,
       LiteLLM proxies, etc.) conventionally expose the native Anthropic
       protocol under a ``/anthropic`` suffix — treat those as
@@ -107,6 +115,12 @@ def _detect_api_mode_for_url(base_url: str) -> Optional[str]:
         return "codex_responses"
     if hostname == "api.openai.com":
         return "codex_responses"
+    # Direct native Anthropic host: realign with providers.determine_api_mode,
+    # which already maps this host to anthropic_messages. The exact-hostname
+    # match rejects lookalike subdomains (api.anthropic.com.attacker.test) and
+    # path-segment spoofing (proxy.test/api.anthropic.com/v1). (#32243)
+    if hostname == "api.anthropic.com":
+        return "anthropic_messages"
     path = urlparse(normalized).path.rstrip("/")
     if path.endswith("/anthropic") or path.endswith("/anthropic/v1"):
         return "anthropic_messages"
@@ -334,6 +348,17 @@ def _parse_api_mode(raw: Any) -> Optional[str]:
     return None
 
 
+def _nous_inference_base_url_override() -> str:
+    """Return the trusted Nous runtime base URL override, if configured.
+
+    Delegates to ``auth._nous_inference_env_override`` so every
+    ``NOUS_INFERENCE_BASE_URL`` read shares one normalization path
+    (trailing-slash stripping, blank → empty). The env source is trusted
+    and intentionally bypasses the network host allowlist there.
+    """
+    return _nous_inference_env_override() or ""
+
+
 def _maybe_apply_codex_app_server_runtime(
     *,
     provider: str,
@@ -412,6 +437,7 @@ def _resolve_runtime_from_pool_entry(
         api_mode = "codex_responses"
     elif provider == "nous":
         api_mode = "chat_completions"
+        base_url = _nous_inference_base_url_override() or base_url
     elif provider == "copilot":
         api_mode = _copilot_runtime_api_mode(model_cfg, getattr(entry, "runtime_api_key", ""))
         base_url = base_url or PROVIDER_REGISTRY["copilot"].inference_base_url
@@ -1359,6 +1385,7 @@ def _resolve_explicit_runtime(
         state = auth_mod.get_provider_auth_state("nous") or {}
         base_url = (
             explicit_base_url
+            or _nous_inference_base_url_override()
             or str(state.get("inference_base_url") or auth_mod.DEFAULT_NOUS_INFERENCE_URL).strip().rstrip("/")
         )
         # Only use the agent_key compatibility field for inference when it
diff --git a/scripts/release.py b/scripts/release.py
index ec347f98aa28..27a6f0084cb2 100755
--- a/scripts/release.py
+++ b/scripts/release.py
@@ -45,7 +45,6 @@
 
 # Auto-extracted from noreply emails + manual overrides
 AUTHOR_MAP = {
-    "5848605+itenev@users.noreply.github.com": "itenev",  # PR #22753 salvage (asyncify model-context resolution in gateway message path so blocking requests.get can't starve Discord heartbeats)
     "290873280+rrevenanttt@users.noreply.github.com": "rrevenanttt",  # PR #40773 salvage (close hardline rm bypass via quoted paths and ${HOME} brace form)
     "290871358+Vesna-9@users.noreply.github.com": "Vesna-9",  # PR #41274 salvage (collapse shell line continuations before dangerous/hardline pattern matching so `rm -rf \<newline>/` can't bypass the yolo-proof hardline floor)
     "214165399+kernel-t1@users.noreply.github.com": "kernel-t1",  # PR #41349 salvage (.env sanitizer: only split when line starts with a known KEY= and preceding values are plain tokens; keep URL/query/whitespace secrets verbatim)
@@ -1204,6 +1203,7 @@
     "xiayh17@gmail.com": "xiayh0107",
     "zhujianxyz@gmail.com": "opriz",
     "tuancanhnguyen706@gmail.com": "xxxigm",
+    "timchris.roth@pm.me": "x9x9x9x9x9x91",
     "larcombe.n@gmail.com": "NickLarcombe",
     "54813621+xxxigm@users.noreply.github.com": "xxxigm",
     "asurla@nvidia.com": "anniesurla",
diff --git a/tests/hermes_cli/test_anthropic_oauth_routes_to_messages_api.py b/tests/hermes_cli/test_anthropic_oauth_routes_to_messages_api.py
new file mode 100644
index 000000000000..e813a375cc5c
--- /dev/null
+++ b/tests/hermes_cli/test_anthropic_oauth_routes_to_messages_api.py
@@ -0,0 +1,205 @@
+"""Regression coverage for issue #32243.
+
+OAuth Pro/Max credentials must always reach Anthropic via the native
+``/v1/messages`` endpoint, never the OpenAI-compat ``/chat/completions``
+shim — the latter bills against a separate "extra usage" pool that
+Pro/Max subscriptions don't fund, so any request that lands on it 400s
+with "You're out of extra usage" the moment the gateway starts.
+
+The root cause was an inconsistency between two URL→api_mode helpers:
+
+* ``hermes_cli.providers.determine_api_mode`` correctly mapped
+  ``api.anthropic.com`` to ``anthropic_messages``.
+* ``hermes_cli.runtime_provider._detect_api_mode_for_url`` did NOT, so
+  every code path that fell back to URL-only detection (named custom
+  providers, direct-alias resolution, the api-key fallback inside
+  ``resolve_runtime_provider``) returned ``None`` for that host and
+  defaulted to ``chat_completions``.
+
+Exhaustive host-shape coverage for the helper itself lives in
+``test_detect_api_mode_for_url.py::TestDirectAnthropicHost``.  The
+tests below pin the **integration contract**: every runtime branch
+that resolves an Anthropic endpoint must return
+``api_mode == "anthropic_messages"``, so a future refactor of any
+single branch cannot silently revert #32243.
+"""
+
+from __future__ import annotations
+
+from hermes_cli import runtime_provider as rp
+
+
+class TestExplicitRuntimeForAnthropic:
+    """``_resolve_explicit_runtime`` with provider='anthropic' must
+    always return ``api_mode='anthropic_messages'`` regardless of
+    base_url shape or stale persisted ``model.api_mode`` values.
+
+    Exercised whenever the user (or a Hermes subcommand) passes an
+    explicit ``--api-key`` / ``--base-url`` override to the runtime
+    resolver.
+    """
+
+    def test_explicit_args_route_to_messages_api(self):
+        result = rp._resolve_explicit_runtime(
+            provider="anthropic",
+            requested_provider="anthropic",
+            model_cfg={},
+            explicit_api_key="sk-ant-oat01-foo",
+            explicit_base_url="https://api.anthropic.com",
+        )
+        assert result is not None
+        assert result["api_mode"] == "anthropic_messages"
+        assert result["provider"] == "anthropic"
+        assert result["base_url"] == "https://api.anthropic.com"
+
+    def test_stale_chat_completions_api_mode_in_config_is_ignored(self):
+        # A user who previously had ``provider: openai`` and switched to
+        # anthropic might still have ``model.api_mode: chat_completions``
+        # in their config.yaml.  The anthropic branch must hard-pin
+        # the mode — Anthropic's chat_completions shim is the bug
+        # locus of #32243 and must never be reachable from this path.
+        result = rp._resolve_explicit_runtime(
+            provider="anthropic",
+            requested_provider="anthropic",
+            model_cfg={"provider": "anthropic", "api_mode": "chat_completions"},
+            explicit_api_key="sk-ant-oat01-foo",
+            explicit_base_url="https://api.anthropic.com",
+        )
+        assert result is not None
+        assert result["api_mode"] == "anthropic_messages"
+
+    def test_no_explicit_args_returns_none(self):
+        # Guard the gating contract — _resolve_explicit_runtime only
+        # fires when an explicit override is present; without one it
+        # must return None so the caller falls through to the pool /
+        # top-level anthropic branch.
+        assert (
+            rp._resolve_explicit_runtime(
+                provider="anthropic",
+                requested_provider="anthropic",
+                model_cfg={"provider": "anthropic"},
+            )
+            is None
+        )
+
+
+class TestPoolEntryForAnthropic:
+    """``_resolve_runtime_from_pool_entry`` is what runs when a user
+    has added an OAuth credential via ``hermes auth add anthropic
+    --type oauth`` (the exact flow from #32243).  Pin the contract
+    alongside the URL-detector test so all three runtime branches
+    stay aligned and a future refactor of one cannot diverge from
+    the others.
+    """
+
+    def test_oauth_pool_entry_routes_to_messages_api(self):
+        class _Entry:
+            access_token = "sk-ant-oat01-pool"
+            runtime_api_key = "sk-ant-oat01-pool"
+            source = "manual:hermes_pkce"
+            base_url = "https://api.anthropic.com"
+
+        resolved = rp._resolve_runtime_from_pool_entry(
+            provider="anthropic",
+            entry=_Entry(),
+            requested_provider="anthropic",
+            model_cfg={"provider": "anthropic"},
+        )
+
+        assert resolved["provider"] == "anthropic"
+        assert resolved["api_mode"] == "anthropic_messages"
+        assert resolved["base_url"] == "https://api.anthropic.com"
+
+    def test_stale_chat_completions_api_mode_in_config_is_ignored(self):
+        # Same regression as the explicit-runtime test above, but on
+        # the pool path: a stale persisted chat_completions api_mode
+        # must NOT override the provider-pin.
+        class _Entry:
+            access_token = "sk-ant-oat01-pool"
+            runtime_api_key = "sk-ant-oat01-pool"
+            source = "manual:hermes_pkce"
+            base_url = "https://api.anthropic.com"
+
+        resolved = rp._resolve_runtime_from_pool_entry(
+            provider="anthropic",
+            entry=_Entry(),
+            requested_provider="anthropic",
+            model_cfg={
+                "provider": "anthropic",
+                "api_mode": "chat_completions",
+            },
+        )
+
+        assert resolved["api_mode"] == "anthropic_messages"
+
+
+class TestCustomProviderUrlFallback:
+    """The detector fix's actual reachable path: a user-defined
+    ``providers:`` / ``custom_providers:`` entry whose ``api`` URL
+    points at ``api.anthropic.com``, with no explicit ``api_mode`` /
+    ``transport`` field.
+
+    Pre-fix: this falls through ``_try_resolve_from_custom_pool`` →
+    ``_detect_api_mode_for_url("https://api.anthropic.com")`` → None →
+    default ``chat_completions`` → request lands on the OpenAI-compat
+    shim → "out of extra usage" 400.
+
+    Post-fix: the detector returns ``anthropic_messages`` so the same
+    config routes to ``/v1/messages`` where Pro/Max OAuth is billed.
+    """
+
+    def test_url_fallback_picks_messages_api(self, monkeypatch):
+        class _Entry:
+            access_token = "sk-ant-oat01-custom-pool"
+            runtime_api_key = "sk-ant-oat01-custom-pool"
+            source = "custom-pool"
+
+        class _Pool:
+            def has_credentials(self):
+                return True
+
+            def select(self):
+                return _Entry()
+
+        monkeypatch.setattr(rp, "get_custom_provider_pool_key", lambda *a, **k: "custom:my-claude")
+        monkeypatch.setattr(rp, "load_pool", lambda key: _Pool())
+
+        resolved = rp._try_resolve_from_custom_pool(
+            "https://api.anthropic.com",
+            "custom",
+        )
+
+        assert resolved is not None
+        assert resolved["api_mode"] == "anthropic_messages"
+
+    def test_explicit_api_mode_override_still_wins(self, monkeypatch):
+        # The detector is only consulted as a fallback — when the
+        # custom-pool caller passes an explicit api_mode (e.g. from a
+        # ``transport: chat_completions`` config entry), that takes
+        # priority.  Pinned so the fix doesn't accidentally hijack a
+        # user who DELIBERATELY pointed a chat_completions transport
+        # at api.anthropic.com (uncommon but valid for OpenAI-compat
+        # experiments).
+        class _Entry:
+            access_token = "k"
+            runtime_api_key = "k"
+            source = "x"
+
+        class _Pool:
+            def has_credentials(self):
+                return True
+
+            def select(self):
+                return _Entry()
+
+        monkeypatch.setattr(rp, "get_custom_provider_pool_key", lambda *a, **k: "custom:my-claude")
+        monkeypatch.setattr(rp, "load_pool", lambda key: _Pool())
+
+        resolved = rp._try_resolve_from_custom_pool(
+            "https://api.anthropic.com",
+            "custom",
+            api_mode_override="chat_completions",
+        )
+
+        assert resolved is not None
+        assert resolved["api_mode"] == "chat_completions"
diff --git a/tests/hermes_cli/test_detect_api_mode_for_url.py b/tests/hermes_cli/test_detect_api_mode_for_url.py
index e9ee41dea71d..e776d0d4e46e 100644
--- a/tests/hermes_cli/test_detect_api_mode_for_url.py
+++ b/tests/hermes_cli/test_detect_api_mode_for_url.py
@@ -1,10 +1,15 @@
 """Tests for hermes_cli.runtime_provider._detect_api_mode_for_url.
 
-The helper maps base URLs to api_modes for three cases:
-  * api.openai.com  → codex_responses
-  * api.x.ai        → codex_responses
-  * */anthropic     → anthropic_messages (third-party gateways like MiniMax,
-                                          Zhipu GLM, LiteLLM proxies)
+The helper maps base URLs to api_modes for four cases:
+  * api.openai.com    → codex_responses
+  * api.x.ai          → codex_responses
+  * api.anthropic.com → anthropic_messages (Pro/Max OAuth is only billed
+                                            against /v1/messages; the
+                                            chat_completions shim counts
+                                            against a separate empty
+                                            "extra usage" pool, see #32243)
+  * */anthropic       → anthropic_messages (third-party gateways like MiniMax,
+                                            Zhipu GLM, LiteLLM proxies)
 
 Consolidating the /anthropic detection in this helper (instead of three
 inline ``endswith`` checks spread across _resolve_runtime_from_pool_entry,
@@ -38,6 +43,49 @@ def test_xai_host_suffix_does_not_match(self):
         assert _detect_api_mode_for_url("https://api.x.ai.example/v1") is None
 
 
+class TestDirectAnthropicHost:
+    """Native api.anthropic.com → /v1/messages. Pinned for issue #32243.
+
+    The Anthropic OpenAI-compat ``/chat/completions`` shim on the same
+    host bills against a separate "extra usage" pool that Pro/Max OAuth
+    subscriptions don't fund, so a fresh OAuth credential 400s with
+    "out of extra usage" the moment a request lands there. The detector
+    must keep ``api.anthropic.com`` on the native Messages API.
+    """
+
+    def test_bare_host(self):
+        assert _detect_api_mode_for_url("https://api.anthropic.com") == "anthropic_messages"
+
+    def test_with_trailing_slash(self):
+        assert _detect_api_mode_for_url("https://api.anthropic.com/") == "anthropic_messages"
+
+    def test_with_v1_suffix(self):
+        # The Anthropic SDK appends /v1/messages itself but the user's
+        # config may persist the /v1 form — must still resolve.
+        assert _detect_api_mode_for_url("https://api.anthropic.com/v1") == "anthropic_messages"
+
+    def test_uppercase_host_tolerated(self):
+        assert _detect_api_mode_for_url("https://API.ANTHROPIC.COM/v1") == "anthropic_messages"
+
+    def test_lookalike_subdomain_does_not_match(self):
+        # ``api.anthropic.com.attacker.test`` is an attacker-controlled
+        # host; the registrable label is ``attacker``, not Anthropic.
+        # Must NOT be routed to anthropic_messages — leaking an
+        # Anthropic OAuth token there is the worst case.
+        assert (
+            _detect_api_mode_for_url("https://api.anthropic.com.attacker.test/v1")
+            is None
+        )
+
+    def test_anthropic_path_segment_does_not_match(self):
+        # A reverse proxy under an unrelated host whose path *contains*
+        # ``api.anthropic.com`` should not be classified as native.
+        assert (
+            _detect_api_mode_for_url("https://proxy.example.test/api.anthropic.com/v1")
+            is None
+        )
+
+
 class TestAnthropicMessagesDetection:
     """Third-party gateways that speak the Anthropic protocol under /anthropic."""
 
diff --git a/tests/run_agent/test_provider_fallback.py b/tests/run_agent/test_provider_fallback.py
index b179cc341cc5..8a0e05c332c6 100644
--- a/tests/run_agent/test_provider_fallback.py
+++ b/tests/run_agent/test_provider_fallback.py
@@ -182,6 +182,35 @@ def test_resolves_key_env_for_fallback_provider(self):
             assert agent._try_activate_fallback() is True
             assert mock_rpc.call_args.kwargs["explicit_api_key"] == "env-secret"
 
+    def test_anthropic_host_custom_provider_uses_anthropic_messages(self):
+        """A custom provider on the native api.anthropic.com host (no
+        "/anthropic" path suffix, name != "anthropic") must resolve to the
+        anthropic_messages wire protocol — not default to chat_completions,
+        which POSTs /v1/chat/completions and 404s. Mirrors the primary-path
+        determine_api_mode() host check."""
+        fbs = [
+            {
+                "provider": "cron-anthropic",
+                "model": "claude-sonnet-4-6",
+                "base_url": "https://api.anthropic.com",
+                "key_env": "MY_FALLBACK_KEY",
+            }
+        ]
+        agent = _make_agent(fallback_model=fbs)
+        with (
+            patch.dict("os.environ", {"MY_FALLBACK_KEY": "env-secret"}, clear=False),
+            patch(
+                "agent.auxiliary_client.resolve_provider_client",
+                return_value=(
+                    _mock_client(base_url="https://api.anthropic.com"),
+                    "claude-sonnet-4-6",
+                ),
+            ),
+            patch("hermes_cli.model_normalize.normalize_model_for_provider", side_effect=lambda m, p: m),
+        ):
+            assert agent._try_activate_fallback() is True
+            assert agent.api_mode == "anthropic_messages"
+
 
 # ── Pool-rotation vs fallback gating (#11314) ────────────────────────────