fix(sync): address generated review findings

scripts/check_deliberate_break.py

@@ -15,6 +15,7 @@
 import tempfile
 from collections.abc import Iterator
 from dataclasses import dataclass
+from importlib import metadata
 from io import BytesIO
 from pathlib import Path
 
@@ -32,7 +33,8 @@
     r"\b(assert|expect\(|pytest\.raises\(|assert\.)\b",
 )
 DEFAULT_TIMEOUT_SECONDS = 120
-PYTEST_RUNTIME_DEPENDENCIES = ("pyyaml==6.0.3",)
+PYTEST_RUNTIME_VERSION = "6.0.3"
+PYTEST_RUNTIME_DEPENDENCIES = (f"pyyaml=={PYTEST_RUNTIME_VERSION}",)
 
 
 @dataclass(frozen=True)
@@ -140,8 +142,10 @@ def _ensure_pytest_runtime_deps() -> None:
     Actions ``action_required`` approval wait on workflow-touching PRs.
     """
     try:
-        import yaml  # noqa: F401
-    except ImportError:
+        installed_version = metadata.version("PyYAML")
+    except metadata.PackageNotFoundError:
+        installed_version = None
+    if installed_version != PYTEST_RUNTIME_VERSION:
         subprocess.run(
             [
                 sys.executable,

scripts/langchain/pr_verifier.py

@@ -685,7 +685,7 @@ def _text_from_response_content(content: object) -> str | None:
             for block in content
             if isinstance(block, dict) and isinstance(block.get("text"), str)
         ]
-        if text_blocks:
+        if any(block.strip() for block in text_blocks):
             # Concatenate without a separator: a provider may split one JSON
             # document across blocks, and an inserted newline inside a string
             # literal would make the reassembled payload invalid JSON.

tests/scripts/test_check_deliberate_break.py

@@ -1,4 +1,3 @@
-import builtins
 import os
 import subprocess
 import sys
@@ -260,20 +259,20 @@ def test_cli_skips_without_marker(tmp_path) -> None:
     assert "skipped: no deliberate-break marker" in completed.stdout
 
 
-def test_runtime_dependency_installer_uses_locked_pyyaml(monkeypatch) -> None:
-    real_import = builtins.__import__
+@pytest.mark.parametrize("installed_version", [None, "6.0.2"])
+def test_runtime_dependency_installer_uses_locked_pyyaml(monkeypatch, installed_version) -> None:
     calls: list[tuple[object, dict[str, object]]] = []
 
-    def missing_yaml(name, *args, **kwargs):
-        if name == "yaml":
-            raise ImportError("PyYAML missing")
-        return real_import(name, *args, **kwargs)
+    def package_version(_name):
+        if installed_version is None:
+            raise deliberate_break.metadata.PackageNotFoundError
+        return installed_version
 
     def record_install(*args, **kwargs):
         calls.append((args, kwargs))
         return subprocess.CompletedProcess(args[0], 0, "", "")
 
-    monkeypatch.setattr(builtins, "__import__", missing_yaml)
+    monkeypatch.setattr(deliberate_break.metadata, "version", package_version)
     monkeypatch.setattr(deliberate_break.subprocess, "run", record_install)
 
     deliberate_break._ensure_pytest_runtime_deps()
@@ -300,6 +299,25 @@ def record_install(*args, **kwargs):
     ]
 
 
+def test_runtime_dependency_installer_accepts_exact_locked_pyyaml(monkeypatch) -> None:
+    calls: list[object] = []
+
+    monkeypatch.setattr(
+        deliberate_break.metadata,
+        "version",
+        lambda _name: deliberate_break.PYTEST_RUNTIME_VERSION,
+    )
+    monkeypatch.setattr(
+        deliberate_break.subprocess,
+        "run",
+        lambda *args, **kwargs: calls.append((args, kwargs)),
+    )
+
+    deliberate_break._ensure_pytest_runtime_deps()
+
+    assert calls == []
+
+
 def _sound_spec(repo: Path) -> tuple[str, object]:
     _write_app(repo, 0)
     base = _commit(repo, "base behavior")

tests/scripts/test_pr_verifier_structured_output.py

@@ -226,6 +226,26 @@ def test_text_from_response_content_concatenates_split_text_blocks_without_separ
     assert pr_verifier._coerce_response_content(blocks) == encoded
 
 
+def test_text_from_response_content_preserves_blank_blocks_between_text() -> None:
+    blocks = [
+        {"type": "text", "text": '{"summary":"not'},
+        {"type": "text", "text": " "},
+        {"type": "text", "text": 'safe"}'},
+    ]
+
+    assert pr_verifier._text_from_response_content(blocks) == '{"summary":"not safe"}'
+
+
+def test_text_from_response_content_ignores_blank_text_blocks() -> None:
+    blocks = [
+        {"type": "text", "text": "  \n"},
+        {"type": "thinking", "signature": "still thinking"},
+    ]
+
+    assert pr_verifier._text_from_response_content(blocks) is None
+    assert pr_verifier._coerce_response_content(blocks) == json.dumps(blocks, default=str)
+
+
 def test_evaluate_pr_valid_output_no_repair(monkeypatch: pytest.MonkeyPatch) -> None:
     payload = _valid_payload()
     good = json.dumps(payload)

tests/tools/test_evaluate_model_benchmark.py

@@ -293,6 +293,17 @@ def test_override_floor_below_one_is_rejected(floor):
         evaluator.evaluate_benchmark(_thin_payload(), policy)
 
 
+@pytest.mark.parametrize("floor", [1.9, "2", None, True])
+def test_override_floor_must_be_an_integer(floor):
+    policy = _policy()
+    policy["profiles"]["verifier-balanced"]["approval_stage"][
+        "minimum_cases_per_category_overrides"
+    ] = {"review-thread-debt": floor}
+
+    with pytest.raises(ValueError, match="must be integers"):
+        evaluator.evaluate_benchmark(_thin_payload(), policy)
+
+
 def test_override_for_unknown_category_is_rejected():
     policy = _policy()
     policy["profiles"]["verifier-balanced"]["approval_stage"][

tools/evaluate_model_benchmark.py

@@ -165,7 +165,11 @@ def evaluate_benchmark(payload: dict[str, Any], policy: dict[str, Any]) -> dict[
     category_floors: dict[str, int] = {}
     for category in required_categories:
         floor = raw_overrides.get(category, minimum_per_category)
-        floor = int(floor)
+        if type(floor) is not int:
+            raise ValueError(
+                "minimum_cases_per_category_overrides values must be integers "
+                f"(got {floor!r} for {category!r})"
+            )
         if floor < 1:
             raise ValueError(
                 "minimum_cases_per_category_overrides values must be >= 1 "