Merge nucleic/plucky-north-vole-sdna into dev

tests/test_label_swe_chat_prompts.py

@@ -64,6 +64,25 @@ class LabelSWEChatPromptsTests(unittest.TestCase):
         encoded = json.dumps(state)
         self.assertNotIn("It fails only on CI.", encoded)
 
+    def test_response_sanitizer_preserves_all_prose_and_removes_code_payloads(self):
+        response = (
+            "I found the likely cause.\n\n"
+            "```swift\nlet secret = \"not teacher context\"\n```\n\n"
+            "The fix is to await the task.\n"
+            "<tool_use>{\"cmd\": \"rm -rf /\"}</tool_use>\n"
+            "Then rerun the focused test."
+        )
+        sanitized = label_swe_chat_prompts.response_for_labeling(response, 2_000)
+        self.assertIn("I found the likely cause.", sanitized)
+        self.assertIn("The fix is to await the task.", sanitized)
+        self.assertIn("Then rerun the focused test.", sanitized)
+        self.assertNotIn("let secret", sanitized)
+        self.assertNotIn("rm -rf", sanitized)
+
+    def test_response_sanitizer_refuses_to_truncate_prose(self):
+        with self.assertRaisesRegex(ValueError, "rather than truncating"):
+            label_swe_chat_prompts.response_for_labeling("x" * 1_001, 1_000)
+
 
 if __name__ == "__main__":
     unittest.main()