Merge nucleic/upbeat-yarn-seal-ekbe into dev

tests/test_label_nucleic_prompts.py

@@ -22,9 +22,10 @@ from pathlib import Path
 
 args = sys.argv[1:]
 model_index = args.index("--model")
-if args[model_index + 1] != "gpt-5.6-terra":
+if args[model_index + 1] != os.environ.get("EXPECTED_MODEL", "gpt-5.6-terra"):
     raise SystemExit("wrong model")
-if 'model_reasoning_effort="low"' not in args:
+expected_effort = os.environ.get("EXPECTED_REASONING_EFFORT", "low")
+if f'model_reasoning_effort="{expected_effort}"' not in args:
     raise SystemExit("wrong reasoning effort")
 
 prompt = sys.stdin.read()
@@ -95,6 +96,12 @@ class LabelNucleicPromptsTests(unittest.TestCase):
     ) -> subprocess.CompletedProcess[str]:
         environment = os.environ.copy()
         environment["FAKE_CODEX_LOG"] = str(root / "calls.log")
+        if "--model" in extra:
+            environment["EXPECTED_MODEL"] = extra[extra.index("--model") + 1]
+        if "--reasoning-effort" in extra:
+            environment["EXPECTED_REASONING_EFFORT"] = extra[
+                extra.index("--reasoning-effort") + 1
+            ]
         return subprocess.run(
             [
                 sys.executable,
@@ -224,6 +231,30 @@ class LabelNucleicPromptsTests(unittest.TestCase):
             )
             self.assertIn("pending=0", second.stdout)
 
+    def test_custom_model_and_reasoning_effort_are_forwarded(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            fake_codex = self.make_fake_codex(root)
+            input_path = root / "unlabeled.jsonl"
+            input_path.write_text(
+                json.dumps({"prompt": "Write the migration note"}) + "\n",
+                encoding="utf-8",
+            )
+
+            result = self.run_script(
+                root,
+                input_path,
+                root / "labeled.jsonl",
+                fake_codex,
+                "--model",
+                "gpt-5.6-sol",
+                "--reasoning-effort",
+                "high",
+            )
+
+            self.assertEqual(0, result.returncode, result.stderr)
+            self.assertIn("model=gpt-5.6-sol reasoning=high", result.stdout)
+
     def test_head_tail_excerpt_is_bounded(self):
         prompt = "HEAD" + ("x" * 4_000) + "TAIL"
         excerpt = label_nucleic_prompts.excerpt_for_labeling(prompt, 1_000)