diff --git a/tests/claude_code/_driver_unit_tests/test_cli_driver.py b/tests/claude_code/_driver_unit_tests/test_cli_driver.py
index fbdea1a0aa4..4766111c08a 100644
--- a/tests/claude_code/_driver_unit_tests/test_cli_driver.py
+++ b/tests/claude_code/_driver_unit_tests/test_cli_driver.py
@@ -18,6 +18,7 @@ import pytest
 from tests.claude_code.cli_driver import (
     ClaudeCLIError,
     DriverResult,
+    failure_diagnostic,
     run_claude,
 )
 
@@ -237,3 +238,102 @@ def test_run_claude_validates_required_params():
             api_key="",
             runner=runner,
         )
+
+
+# ---------------------------------------------------------------------------
+# failure_diagnostic
+#
+# Regression coverage for the bring-up incident where the proxy was started
+# with the wrong config and tests reported only `claude CLI exited 1` while
+# the actual 400 from LiteLLM was sitting in stdout. The helper must surface
+# api_status, the assistant text (where API errors land), stderr, and the
+# exit code together — and gracefully degrade when individual pieces are
+# missing.
+# ---------------------------------------------------------------------------
+
+
+def test_failure_diagnostic_surfaces_api_error_text_from_stdout():
+    """The CLI hides 4xx/5xx from the proxy in `assistant.message.content` text."""
+    api_error_text = (
+        'API Error: 400 {"error":{"message":"litellm.BadRequestError: '
+        "You passed in model=claude-haiku-4-5. There are no healthy "
+        'deployments..."}}'
+    )
+    result = DriverResult(
+        text=api_error_text,
+        events=[
+            {
+                "type": "assistant",
+                "message": {"content": [{"type": "text", "text": api_error_text}]},
+            },
+            {
+                "type": "result",
+                "is_error": True,
+                "api_error_status": 400,
+                "result": api_error_text,
+            },
+        ],
+        exit_code=1,
+        stderr="",
+    )
+
+    diag = failure_diagnostic(result)
+
+    assert "exit=1" in diag
+    assert "api_status=400" in diag
+    assert "There are no healthy deployments" in diag
+
+
+def test_failure_diagnostic_falls_back_to_stderr_when_no_text():
+    result = DriverResult(text="", events=[], exit_code=2, stderr="boom\n")
+    diag = failure_diagnostic(result)
+    assert "exit=2" in diag
+    assert "stderr=boom" in diag
+
+
+def test_failure_diagnostic_handles_completely_empty_result():
+    """A run that produced literally nothing should still yield a useful string."""
+    result = DriverResult(text="", events=[], exit_code=137, stderr="")
+    diag = failure_diagnostic(result)
+    assert "exit=137" in diag
+    assert "no diagnostic output" in diag
+
+
+def test_failure_diagnostic_truncates_long_text():
+    """Don't let a 5MB HTML 502 page from a load balancer wreck the matrix JSON."""
+    huge = "x" * 5000
+    result = DriverResult(text=huge, events=[], exit_code=1, stderr="")
+    diag = failure_diagnostic(result, max_len=100)
+    assert "truncated" in diag
+    # Allow some slack for the prefix/suffix/separator characters.
+    assert len(diag) < 300
+
+
+def test_failure_diagnostic_ignores_non_int_api_error_status():
+    """The CLI sometimes emits api_error_status as a string; don't crash."""
+    result = DriverResult(
+        text="oops",
+        events=[{"type": "result", "api_error_status": "n/a"}],
+        exit_code=1,
+        stderr="",
+    )
+    diag = failure_diagnostic(result)
+    assert "api_status" not in diag
+    assert "text=oops" in diag
+
+
+def test_failure_diagnostic_uses_last_result_event_status():
+    """If multiple `result` events appear, the most recent status wins."""
+    result = DriverResult(
+        text="",
+        events=[
+            {"type": "result", "api_error_status": 500},
+            {"type": "assistant", "message": {"content": []}},
+            {"type": "result", "api_error_status": 429},
+        ],
+        exit_code=1,
+        stderr="",
+    )
+    diag = failure_diagnostic(result)
+    assert "api_status=429" in diag
+    assert "500" not in diag
diff --git a/tests/claude_code/basic_messaging_non_streaming/test_anthropic.py b/tests/claude_code/basic_messaging_non_streaming/test_anthropic.py
index 7e0a1f16583..ee6314d5ded 100644
--- a/tests/claude_code/basic_messaging_non_streaming/test_anthropic.py
+++ b/tests/claude_code/basic_messaging_non_streaming/test_anthropic.py
@@ -23,7 +23,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -63,6 +63,9 @@ def test_basic_messaging_non_streaming_anthropic(compat_result, model):
             f"{PROXY_BASE_URL_ENV} / {PROXY_API_KEY_ENV} not configured", pytrace=False
         )
 
+    print(f"base_url: {base_url}")
+    print(f"api_key: {api_key}")
+
     try:
         result = run_claude(
             prompt="Reply with the single word 'pong' and nothing else.",
@@ -79,10 +82,12 @@ def test_basic_messaging_non_streaming_anthropic(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/basic_messaging_non_streaming/test_azure.py b/tests/claude_code/basic_messaging_non_streaming/test_azure.py
index 22b09d24bc5..43e10f65d66 100644
--- a/tests/claude_code/basic_messaging_non_streaming/test_azure.py
+++ b/tests/claude_code/basic_messaging_non_streaming/test_azure.py
@@ -29,7 +29,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -86,10 +86,12 @@ def test_basic_messaging_non_streaming_azure(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/basic_messaging_non_streaming/test_bedrock_converse.py b/tests/claude_code/basic_messaging_non_streaming/test_bedrock_converse.py
index dc604f65acb..dca60237a28 100644
--- a/tests/claude_code/basic_messaging_non_streaming/test_bedrock_converse.py
+++ b/tests/claude_code/basic_messaging_non_streaming/test_bedrock_converse.py
@@ -24,7 +24,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -75,10 +75,12 @@ def test_basic_messaging_non_streaming_bedrock_converse(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/basic_messaging_non_streaming/test_bedrock_invoke.py b/tests/claude_code/basic_messaging_non_streaming/test_bedrock_invoke.py
index 8dc4ef9fff0..e6bd347c167 100644
--- a/tests/claude_code/basic_messaging_non_streaming/test_bedrock_invoke.py
+++ b/tests/claude_code/basic_messaging_non_streaming/test_bedrock_invoke.py
@@ -24,7 +24,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -75,10 +75,12 @@ def test_basic_messaging_non_streaming_bedrock_invoke(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/basic_messaging_non_streaming/test_vertex_ai.py b/tests/claude_code/basic_messaging_non_streaming/test_vertex_ai.py
index a32d8d720ff..df6d79c710d 100644
--- a/tests/claude_code/basic_messaging_non_streaming/test_vertex_ai.py
+++ b/tests/claude_code/basic_messaging_non_streaming/test_vertex_ai.py
@@ -24,7 +24,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -75,10 +75,12 @@ def test_basic_messaging_non_streaming_vertex_ai(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/basic_messaging_streaming/test_anthropic.py b/tests/claude_code/basic_messaging_streaming/test_anthropic.py
index 7a7ebfbe3db..ec52a24b5ae 100644
--- a/tests/claude_code/basic_messaging_streaming/test_anthropic.py
+++ b/tests/claude_code/basic_messaging_streaming/test_anthropic.py
@@ -25,7 +25,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -74,10 +74,13 @@ def test_basic_messaging_streaming_anthropic(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.events:
diff --git a/tests/claude_code/basic_messaging_streaming/test_azure.py b/tests/claude_code/basic_messaging_streaming/test_azure.py
index 96c10439b54..3b9d78b4912 100644
--- a/tests/claude_code/basic_messaging_streaming/test_azure.py
+++ b/tests/claude_code/basic_messaging_streaming/test_azure.py
@@ -23,7 +23,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -72,10 +72,12 @@ def test_basic_messaging_streaming_azure(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.events:
diff --git a/tests/claude_code/basic_messaging_streaming/test_bedrock_converse.py b/tests/claude_code/basic_messaging_streaming/test_bedrock_converse.py
index c3b1bcdf52c..26008ed482b 100644
--- a/tests/claude_code/basic_messaging_streaming/test_bedrock_converse.py
+++ b/tests/claude_code/basic_messaging_streaming/test_bedrock_converse.py
@@ -19,7 +19,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -68,10 +68,12 @@ def test_basic_messaging_streaming_bedrock_converse(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.events:
diff --git a/tests/claude_code/basic_messaging_streaming/test_bedrock_invoke.py b/tests/claude_code/basic_messaging_streaming/test_bedrock_invoke.py
index 6f0aa0fec0b..0d0d1ee9a33 100644
--- a/tests/claude_code/basic_messaging_streaming/test_bedrock_invoke.py
+++ b/tests/claude_code/basic_messaging_streaming/test_bedrock_invoke.py
@@ -19,7 +19,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -68,10 +68,12 @@ def test_basic_messaging_streaming_bedrock_invoke(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.events:
diff --git a/tests/claude_code/basic_messaging_streaming/test_vertex_ai.py b/tests/claude_code/basic_messaging_streaming/test_vertex_ai.py
index b2c914e2917..d909547673b 100644
--- a/tests/claude_code/basic_messaging_streaming/test_vertex_ai.py
+++ b/tests/claude_code/basic_messaging_streaming/test_vertex_ai.py
@@ -19,7 +19,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -68,10 +68,12 @@ def test_basic_messaging_streaming_vertex_ai(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.events:
diff --git a/tests/claude_code/cli_driver.py b/tests/claude_code/cli_driver.py
index cd8732491d3..c15b38bcb72 100644
--- a/tests/claude_code/cli_driver.py
+++ b/tests/claude_code/cli_driver.py
@@ -173,6 +173,73 @@ def _extract_assistant_text(events: Sequence[Mapping[str, Any]]) -> str:
     return "".join(chunks)
 
 
+def failure_diagnostic(result: "DriverResult", *, max_len: int = 800) -> str:
+    """Build a human-readable error string from a non-zero `claude` CLI run.
+
+    The CLI is annoying to debug because the most useful failure signal
+    rarely lands on stderr. When the proxy returns an HTTP error, the CLI
+    swallows it into an `assistant`/`result` event on **stdout** with
+    `is_error: true` and a JSON-shaped `text` block — and exits non-zero.
+    Tests that only print `stderr.strip()` see an empty string, which is
+    exactly the situation that masked a misconfigured proxy in early
+    bring-up of the compat matrix.
+
+    This helper concatenates the most useful diagnostic we can find, in
+    priority order:
+
+      1. `result.text` (the assistant's user-visible reply, which is where
+         API errors land in stream-json mode), trimmed
+      2. `api_error_status` from any `result` event, if present
+      3. `result.stderr`, trimmed
+      4. `<no diagnostic output>` as a last resort
+
+    The output is truncated to `max_len` characters so a giant HTML 502
+    page from a misbehaving load balancer doesn't blow up the matrix
+    JSON.
+    """
+    pieces: List[str] = [f"exit={result.exit_code}"]
+
+    # api_error_status only appears on the final `result` event when the
+    # CLI received an HTTP error from the upstream API. Surfacing it
+    # explicitly makes "is this a proxy/auth problem or a CLI problem?"
+    # answerable without re-reading the events list.
+    api_status = _extract_api_error_status(result.events)
+    if api_status is not None:
+        pieces.append(f"api_status={api_status}")
+
+    text = (result.text or "").strip()
+    if text:
+        pieces.append(f"text={_truncate(text, max_len)}")
+
+    stderr = (result.stderr or "").strip()
+    if stderr:
+        pieces.append(f"stderr={_truncate(stderr, max_len)}")
+
+    if len(pieces) == 1:
+        pieces.append("(no diagnostic output)")
+
+    return "; ".join(pieces)
+
+
+def _extract_api_error_status(
+    events: Sequence[Mapping[str, Any]],
+) -> Optional[int]:
+    """Return the `api_error_status` from the last `result` event, if any."""
+    for event in reversed(list(events)):
+        if event.get("type") != "result":
+            continue
+        status = event.get("api_error_status")
+        if isinstance(status, int):
+            return status
+    return None
+
+
+def _truncate(s: str, max_len: int) -> str:
+    if len(s) <= max_len:
+        return s
+    return s[:max_len] + "...(truncated)"
+
+
 def _extract_usage(events: Sequence[Mapping[str, Any]]) -> Optional[Dict[str, Any]]:
     """Return the most recent `usage` block seen on any event, if any.
 
diff --git a/tests/claude_code/extended_thinking/test_anthropic.py b/tests/claude_code/extended_thinking/test_anthropic.py
index 39511f484e9..b9d67f887a0 100644
--- a/tests/claude_code/extended_thinking/test_anthropic.py
+++ b/tests/claude_code/extended_thinking/test_anthropic.py
@@ -21,7 +21,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -95,10 +95,12 @@ def test_extended_thinking_anthropic(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_thinking_block(result.events):
diff --git a/tests/claude_code/extended_thinking/test_azure.py b/tests/claude_code/extended_thinking/test_azure.py
index 6d5e89bfcfa..a03fdcf4b76 100644
--- a/tests/claude_code/extended_thinking/test_azure.py
+++ b/tests/claude_code/extended_thinking/test_azure.py
@@ -28,7 +28,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -97,10 +97,12 @@ def test_extended_thinking_azure(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_thinking_block(result.events):
diff --git a/tests/claude_code/extended_thinking/test_bedrock_converse.py b/tests/claude_code/extended_thinking/test_bedrock_converse.py
index a6941a90afe..de33f343a8b 100644
--- a/tests/claude_code/extended_thinking/test_bedrock_converse.py
+++ b/tests/claude_code/extended_thinking/test_bedrock_converse.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -89,10 +89,12 @@ def test_extended_thinking_bedrock_converse(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_thinking_block(result.events):
diff --git a/tests/claude_code/extended_thinking/test_bedrock_invoke.py b/tests/claude_code/extended_thinking/test_bedrock_invoke.py
index 7fd67aa8efd..27b502207f4 100644
--- a/tests/claude_code/extended_thinking/test_bedrock_invoke.py
+++ b/tests/claude_code/extended_thinking/test_bedrock_invoke.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -89,10 +89,12 @@ def test_extended_thinking_bedrock_invoke(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_thinking_block(result.events):
diff --git a/tests/claude_code/extended_thinking/test_vertex_ai.py b/tests/claude_code/extended_thinking/test_vertex_ai.py
index ee36470d4b0..55dae72ccde 100644
--- a/tests/claude_code/extended_thinking/test_vertex_ai.py
+++ b/tests/claude_code/extended_thinking/test_vertex_ai.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -89,10 +89,12 @@ def test_extended_thinking_vertex_ai(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_thinking_block(result.events):
diff --git a/tests/claude_code/prompt_caching_5m/test_anthropic.py b/tests/claude_code/prompt_caching_5m/test_anthropic.py
index d5b5f75fe3f..c8e450198b3 100644
--- a/tests/claude_code/prompt_caching_5m/test_anthropic.py
+++ b/tests/claude_code/prompt_caching_5m/test_anthropic.py
@@ -27,7 +27,7 @@ from typing import Any, Mapping, Optional
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -88,10 +88,12 @@ def test_prompt_caching_5m_anthropic(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if _cache_tokens(result.usage) <= 0:
diff --git a/tests/claude_code/prompt_caching_5m/test_azure.py b/tests/claude_code/prompt_caching_5m/test_azure.py
index f470d950514..0fc12771997 100644
--- a/tests/claude_code/prompt_caching_5m/test_azure.py
+++ b/tests/claude_code/prompt_caching_5m/test_azure.py
@@ -27,7 +27,7 @@ from typing import Any, Mapping, Optional
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -86,10 +86,12 @@ def test_prompt_caching_5m_azure(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if _cache_tokens(result.usage) <= 0:
diff --git a/tests/claude_code/prompt_caching_5m/test_bedrock_converse.py b/tests/claude_code/prompt_caching_5m/test_bedrock_converse.py
index 417ea9b10ab..782a96c2ace 100644
--- a/tests/claude_code/prompt_caching_5m/test_bedrock_converse.py
+++ b/tests/claude_code/prompt_caching_5m/test_bedrock_converse.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Optional
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -79,10 +79,12 @@ def test_prompt_caching_5m_bedrock_converse(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if _cache_tokens(result.usage) <= 0:
diff --git a/tests/claude_code/prompt_caching_5m/test_bedrock_invoke.py b/tests/claude_code/prompt_caching_5m/test_bedrock_invoke.py
index c70d4ed3e32..5fe6679e7ca 100644
--- a/tests/claude_code/prompt_caching_5m/test_bedrock_invoke.py
+++ b/tests/claude_code/prompt_caching_5m/test_bedrock_invoke.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Optional
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -79,10 +79,12 @@ def test_prompt_caching_5m_bedrock_invoke(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if _cache_tokens(result.usage) <= 0:
diff --git a/tests/claude_code/prompt_caching_5m/test_vertex_ai.py b/tests/claude_code/prompt_caching_5m/test_vertex_ai.py
index bf33d89fffe..f2665979bf1 100644
--- a/tests/claude_code/prompt_caching_5m/test_vertex_ai.py
+++ b/tests/claude_code/prompt_caching_5m/test_vertex_ai.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Optional
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -79,10 +79,12 @@ def test_prompt_caching_5m_vertex_ai(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if _cache_tokens(result.usage) <= 0:
diff --git a/tests/claude_code/test_config.yaml b/tests/claude_code/test_config.yaml
index 05716fa41ca..2943f301fab 100644
--- a/tests/claude_code/test_config.yaml
+++ b/tests/claude_code/test_config.yaml
@@ -76,12 +76,6 @@ model_list:
       vertex_ai_location: us-east5
 
   # ---- Microsoft Foundry (Anthropic deployments on Azure) ----
-  # Anthropic announced Claude Haiku 4.5, Sonnet 4.5/4.6, and Opus 4.1/4.6/4.7
-  # in Microsoft Foundry on 2025-11-18. Foundry exposes these models on an
-  # Anthropic-shape `/anthropic/v1/messages` endpoint (not Azure OpenAI's
-  # chat-completions route), and LiteLLM routes them via the `azure_ai/`
-  # provider prefix. Set AZURE_FOUNDRY_API_BASE to the host (or host +
-  # `/anthropic`); LiteLLM normalizes the URL to `…/anthropic/v1/messages`.
   - model_name: claude-haiku-4-5-azure
     litellm_params:
       model: azure_ai/claude-haiku-4-5
diff --git a/tests/claude_code/tool_use/test_anthropic.py b/tests/claude_code/tool_use/test_anthropic.py
index d2d40654a13..6a0627f0233 100644
--- a/tests/claude_code/tool_use/test_anthropic.py
+++ b/tests/claude_code/tool_use/test_anthropic.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -93,10 +93,12 @@ def test_tool_use_anthropic(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_tool_use_event(result.events):
diff --git a/tests/claude_code/tool_use/test_azure.py b/tests/claude_code/tool_use/test_azure.py
index 1103adc610d..133e9d2b48d 100644
--- a/tests/claude_code/tool_use/test_azure.py
+++ b/tests/claude_code/tool_use/test_azure.py
@@ -24,7 +24,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -92,10 +92,12 @@ def test_tool_use_azure(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_tool_use_event(result.events):
diff --git a/tests/claude_code/tool_use/test_bedrock_converse.py b/tests/claude_code/tool_use/test_bedrock_converse.py
index 9a99c877d59..81b25f76668 100644
--- a/tests/claude_code/tool_use/test_bedrock_converse.py
+++ b/tests/claude_code/tool_use/test_bedrock_converse.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -88,10 +88,12 @@ def test_tool_use_bedrock_converse(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_tool_use_event(result.events):
diff --git a/tests/claude_code/tool_use/test_bedrock_invoke.py b/tests/claude_code/tool_use/test_bedrock_invoke.py
index 9bc9b66a81c..d54776b9bab 100644
--- a/tests/claude_code/tool_use/test_bedrock_invoke.py
+++ b/tests/claude_code/tool_use/test_bedrock_invoke.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -88,10 +88,12 @@ def test_tool_use_bedrock_invoke(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_tool_use_event(result.events):
diff --git a/tests/claude_code/tool_use/test_vertex_ai.py b/tests/claude_code/tool_use/test_vertex_ai.py
index dcfe8d002e5..20c93e54fed 100644
--- a/tests/claude_code/tool_use/test_vertex_ai.py
+++ b/tests/claude_code/tool_use/test_vertex_ai.py
@@ -20,7 +20,7 @@ from typing import Any, Mapping, Sequence
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -88,10 +88,12 @@ def test_tool_use_vertex_ai(compat_result, model):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not _has_tool_use_event(result.events):
diff --git a/tests/claude_code/vision/test_anthropic.py b/tests/claude_code/vision/test_anthropic.py
index a381f9ed018..e0d178fb0ce 100644
--- a/tests/claude_code/vision/test_anthropic.py
+++ b/tests/claude_code/vision/test_anthropic.py
@@ -21,7 +21,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -78,10 +78,12 @@ def test_vision_anthropic(compat_result, model, tmp_path):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/vision/test_azure.py b/tests/claude_code/vision/test_azure.py
index c24c41c05ac..f5e5c249036 100644
--- a/tests/claude_code/vision/test_azure.py
+++ b/tests/claude_code/vision/test_azure.py
@@ -25,7 +25,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -79,10 +79,12 @@ def test_vision_azure(compat_result, model, tmp_path):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/vision/test_bedrock_converse.py b/tests/claude_code/vision/test_bedrock_converse.py
index 50067e9c594..2bdfa1196fd 100644
--- a/tests/claude_code/vision/test_bedrock_converse.py
+++ b/tests/claude_code/vision/test_bedrock_converse.py
@@ -20,7 +20,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -74,10 +74,12 @@ def test_vision_bedrock_converse(compat_result, model, tmp_path):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/vision/test_bedrock_invoke.py b/tests/claude_code/vision/test_bedrock_invoke.py
index e0032e24145..ec9a4287cc3 100644
--- a/tests/claude_code/vision/test_bedrock_invoke.py
+++ b/tests/claude_code/vision/test_bedrock_invoke.py
@@ -20,7 +20,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -74,10 +74,12 @@ def test_vision_bedrock_invoke(compat_result, model, tmp_path):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():
diff --git a/tests/claude_code/vision/test_vertex_ai.py b/tests/claude_code/vision/test_vertex_ai.py
index 7e1101e12ce..2739206dd08 100644
--- a/tests/claude_code/vision/test_vertex_ai.py
+++ b/tests/claude_code/vision/test_vertex_ai.py
@@ -20,7 +20,7 @@ import os
 
 import pytest
 
-from tests.claude_code.cli_driver import ClaudeCLIError, run_claude
+from tests.claude_code.cli_driver import ClaudeCLIError, failure_diagnostic, run_claude
 
 PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
 PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
@@ -74,10 +74,12 @@ def test_vision_vertex_ai(compat_result, model, tmp_path):
         compat_result.set(
             {
                 "status": "fail",
-                "error": f"[{model}] claude CLI exited {result.exit_code}: {result.stderr.strip()}",
+                "error": f"[{model}] claude CLI failed: {failure_diagnostic(result)}",
             }
         )
-        pytest.fail(f"claude CLI exited {result.exit_code} for {model}", pytrace=False)
+        pytest.fail(
+            f"[{model}] claude CLI failed: {failure_diagnostic(result)}", pytrace=False
+        )
         return
 
     if not result.text.strip():