fix(e2e/claude_code): unblock stage collection and align proxy env names with the rest of tests/e2e

tests/e2e/CLAUDE.md

@@ -171,3 +171,16 @@ other.<area>.<case>.<assertion>
   e.g.  other.auth.jwt.valid_token_allows
         other.lifecycle.readiness.reports_db
 ```
+
+## Hard Rules
+- no monkeypatching, mock tests or unit tests of any kind. if a contributor asks you to write an end to end test, do NOT stage a unit test with it. if you find a product gap, call it out in the PR description
+
+- use model management endpoints to create new models for a test. this could be in a conftest / inline for each test. ask the user what they want.
+
+- do not overengineer a test, i need you to write readable, clean code of what would look like a natural user scenario
+
+- when it comes to typing an input schema for an api endpoint, have it type X = A | B | C ... where X = exhaustive union of all supported input schemas and A, B, C typically are composed by a base type. types are only pretty for a api request / response body. make sure to compose types instead of repeating the same base attributes over and over again.
+
+- use the docker-compose to your advantage and spin up a local proxy, make sure all tests pass. if a test fails due to an internally found issue, let users know to create a linear ticket for it. 
+
+- do not use xfail markers, tests should be written in a form that the end user expects it to pass

tests/e2e/claude_code/_basic_messaging.py

@@ -27,19 +27,20 @@
 
 from __future__ import annotations
 
-import os
-from typing import Any, Mapping, Sequence
+from typing import Any, Callable, Mapping, Sequence
 
 import pytest
 
+from claude_code._env import require_proxy
 from claude_code.cli_driver import (
     ClaudeCLIError,
+    DriverResult,
     failure_diagnostic,
     run_claude_models_parallel,
 )
 
-PROXY_BASE_URL_ENV = "LITELLM_PROXY_BASE_URL"
-PROXY_API_KEY_ENV = "LITELLM_PROXY_API_KEY"
+
+ClaudeRunner = Callable[..., Mapping[str, DriverResult | ClaudeCLIError]]
 
 # Floor on the number of `stream_event` records (with delta payloads)
 # we expect to see when the proxy actually streams. With
@@ -79,6 +80,8 @@ def run_basic_messaging_cell(
     models: Sequence[str],
     prompt: str,
     verify_streaming: bool = False,
+    env: Mapping[str, str] | None = None,
+    runner: ClaudeRunner = run_claude_models_parallel,
 ) -> None:
     """Run the shared `basic_messaging_*` × <provider> cell body.
 
@@ -99,28 +102,13 @@ def run_basic_messaging_cell(
     streamed reply to a single ``assistant`` event in
     ``--print --output-format stream-json`` mode).
     """
-    base_url = os.environ.get(PROXY_BASE_URL_ENV)
-    api_key = os.environ.get(PROXY_API_KEY_ENV)
-    if not base_url or not api_key:
-        compat_result.set(
-            {
-                "status": "fail",
-                "error": (
-                    f"missing required env: set {PROXY_BASE_URL_ENV} and "
-                    f"{PROXY_API_KEY_ENV} to point at a running LiteLLM proxy"
-                ),
-            }
-        )
-        pytest.fail(
-            f"{PROXY_BASE_URL_ENV} / {PROXY_API_KEY_ENV} not configured",
-            pytrace=False,
-        )
+    base_url, api_key = require_proxy(compat_result, env=env)
 
     extra_args: Sequence[str] = (
         ("--include-partial-messages",) if verify_streaming else ()
     )
 
-    outcomes = run_claude_models_parallel(
+    outcomes = runner(
         models=models,
         prompt=prompt,
         base_url=base_url,

tests/e2e/claude_code/_builder_unit_tests/fixtures/expected_matrix.json

@@ -26,7 +26,7 @@
       "providers": {
         "anthropic": {
           "status": "fail",
-          "error": "[claude-sonnet-4-6] tool call dropped"
+          "error": "[claude-sonnet-4-5] tool call dropped"
         },
         "bedrock_invoke": {
           "status": "not_applicable",

tests/e2e/claude_code/_builder_unit_tests/fixtures/results.json

@@ -10,7 +10,7 @@
     {
       "feature_id": "basic_messaging_non_streaming",
       "provider": "anthropic",
-      "nodeid": "tests/e2e/claude_code/basic_messaging_non_streaming/test_anthropic.py::test_basic_messaging_non_streaming_anthropic[claude-sonnet-4-6]",
+      "nodeid": "tests/e2e/claude_code/basic_messaging_non_streaming/test_anthropic.py::test_basic_messaging_non_streaming_anthropic[claude-sonnet-4-5]",
       "result": {"status": "pass"}
     },
     {
@@ -28,8 +28,8 @@
     {
       "feature_id": "tool_use",
       "provider": "anthropic",
-      "nodeid": "tests/e2e/claude_code/tool_use/test_anthropic.py::test_x[claude-sonnet-4-6]",
-      "result": {"status": "fail", "error": "[claude-sonnet-4-6] tool call dropped"}
+      "nodeid": "tests/e2e/claude_code/tool_use/test_anthropic.py::test_x[claude-sonnet-4-5]",
+      "result": {"status": "fail", "error": "[claude-sonnet-4-5] tool call dropped"}
     },
     {
       "feature_id": "tool_use",

tests/e2e/claude_code/_builder_unit_tests/test_matrix_builder.py

@@ -393,7 +393,7 @@ def test_build_matrix_6x5_grid_matches_published_sample():
 
     feature_ids = [feature["id"] for feature in manifest["features"]]
     providers = manifest["providers"]
-    models = ["claude-haiku-4-5", "claude-sonnet-4-6", "claude-opus-4-7"]
+    models = ["claude-haiku-4-5", "claude-sonnet-4-5", "claude-opus-4-7"]
 
     results = []
     for feature_id in feature_ids:

tests/e2e/claude_code/_builder_unit_tests/test_v0_layout.py

@@ -152,7 +152,7 @@ def test_per_provider_test_file_imports_and_parametrizes_three_models(
     `claude-opus-4-7-bedrock-invoke`), so we check for the tier
     substrings rather than exact alias names."""
     text = (REPO_ROOT / feature_id / f"test_{provider}.py").read_text()
-    for tier in ("haiku-4-5", "sonnet-4-6", "opus-4-7"):
+    for tier in ("haiku-4-5", "sonnet-4-5", "opus-4-7"):
         assert (
             tier in text
         ), f"{feature_id}/test_{provider}.py does not reference {tier}"

tests/e2e/claude_code/_compat_models.py

@@ -0,0 +1,86 @@
+"""Load the claude_code compat matrix's deployment list from
+``test_config.yaml``.
+
+``test_config.yaml`` is the ground-truth config the stage deployment
+uses; parsing it at fixture time means a change there (new tier, tier
+retirement, provider swap, endpoint rename) reaches the fixture with
+no extra edit. A drift-check test asserts every ``*_MODELS`` list
+referenced by the compat cells is covered by the yaml, so a cell that
+adds a probe for a name the yaml doesn't know about fails loudly at
+collection instead of at 400-time.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Callable, Mapping
+
+import yaml
+
+from models import LiteLLMParamsBody
+
+CONFIG_PATH = Path(__file__).resolve().parent / "test_config.yaml"
+
+
+@dataclass(frozen=True, slots=True)
+class CompatDeployment:
+    model_name: str
+    litellm_params: LiteLLMParamsBody
+
+
+# The yaml uses ``vertex_ai_*`` for the vertex project/location fields
+# (that is the spelling the proxy config file historically standardized
+# on), while ``LiteLLMParamsBody`` names them without the ``_ai`` infix
+# (matching the proxy's DB column). Both spellings resolve at call time
+# on the proxy side, but pydantic silently drops unknown fields, so a
+# raw ``LiteLLMParamsBody(**entry)`` would produce a body with the
+# vertex project stripped - the resulting deployment 400s at
+# ``/v1/messages`` with "Invalid model name". Normalize the yaml keys
+# to the pydantic names in one place.
+_YAML_TO_PYDANTIC_ALIASES = {
+    "vertex_ai_project": "vertex_project",
+    "vertex_ai_location": "vertex_location",
+    "vertex_ai_credentials": "vertex_credentials",
+}
+
+
+def _normalize_params(raw: Mapping[str, object]) -> dict[str, object]:
+    return {_YAML_TO_PYDANTIC_ALIASES.get(k, k): v for k, v in raw.items()}
+
+
+ConfigReader = Callable[[Path], str]
+
+
+def _default_reader(path: Path) -> str:
+    return path.read_text()
+
+
+def load_all_deployments(
+    config_path: Path = CONFIG_PATH,
+    reader: ConfigReader = _default_reader,
+) -> tuple[CompatDeployment, ...]:
+    """Every deployment declared in the yaml, in file order."""
+    doc = yaml.safe_load(reader(config_path))
+    model_list = doc.get("model_list") or []
+    return tuple(
+        CompatDeployment(
+            model_name=entry["model_name"],
+            litellm_params=LiteLLMParamsBody(
+                **_normalize_params(entry["litellm_params"])
+            ),
+        )
+        for entry in model_list
+    )
+
+
+def all_expected_model_names(
+    *,
+    config_path: Path = CONFIG_PATH,
+    reader: ConfigReader = _default_reader,
+) -> frozenset[str]:
+    """Every virtual name the compat matrix declares - the ground truth
+    the cells are supposed to probe. Used by the drift-check test."""
+    return frozenset(
+        d.model_name for d in load_all_deployments(config_path, reader)
+    )