diff --git a/agent/auxiliary_client.py b/agent/auxiliary_client.py
index 117c3e5fda..8d293f02de 100644
--- a/agent/auxiliary_client.py
+++ b/agent/auxiliary_client.py
@@ -5569,6 +5569,13 @@ def _get_cached_client(
         provider, model, async_mode, explicit_base_url=base_url, explicit_api_key=effective_api_key,
         api_mode=api_mode, main_runtime=runtime, is_vision=is_vision, task=task,
     )
+    if client is not None and _aux_probe_active():
+        # Availability probes answer "resolvable?" and must leave the cache untouched: the
+        # probe stub (bare, or wrapped in a Codex/Anthropic adapter whose leaf is the stub)
+        # shares the runtime key, and a cached one is served to every later caller — the
+        # next probe dies in _compat_model() on stub attribute access, so check_fns flip to
+        # False and vision tools vanish for the process lifetime (#87654).
+        return client, model or default_model
     if client is not None:
         with _client_cache_lock:
             if cache_key not in _client_cache:
diff --git a/tests/tools/test_startup_latency_regressions.py b/tests/tools/test_startup_latency_regressions.py
index 45231f94db..439dd1083a 100644
--- a/tests/tools/test_startup_latency_regressions.py
+++ b/tests/tools/test_startup_latency_regressions.py
@@ -34,6 +34,41 @@ class TestAuxProbeMode:
         with aux._client_cache_lock:
             assert key not in aux._client_cache
 
+    def test_probe_stub_never_cached_via_get_cached_client(self):
+        """The inline store in _get_cached_client must honour the same guard.
+
+        test_probe_stub_never_cached above covers _store_cached_client(), but
+        _get_cached_client() assigns to _client_cache directly instead of
+        calling it. That is the path check_fns actually take inside
+        aux_probe_mode(), so without a guard there the probe poisons the entry
+        and the next runtime caller with the same key gets a dead client.
+        """
+        import agent.auxiliary_client as aux
+
+        stub = aux._AuxProbeClientStub()
+        try:
+            with patch.object(
+                aux, "resolve_provider_client", lambda *a, **k: (stub, "m")
+            ):
+                with aux.aux_probe_mode():
+                    client, _model = aux._get_cached_client("probe-guard-test", "m")
+            assert client is stub
+            with aux._client_cache_lock:
+                poisoned = [
+                    key
+                    for key, entry in aux._client_cache.items()
+                    if isinstance(entry[0], aux._AuxProbeClientStub)
+                ]
+            assert poisoned == [], f"probe stub reached the cache: {poisoned}"
+        finally:
+            with aux._client_cache_lock:
+                for key in [
+                    k
+                    for k, e in aux._client_cache.items()
+                    if isinstance(e[0], aux._AuxProbeClientStub)
+                ]:
+                    del aux._client_cache[key]
+
     def test_probe_stub_raises_on_runtime_use(self):
         import agent.auxiliary_client as aux