feat: implement t-604 keyword prompt moderation

apps/moderation/tests.py

@@ -0,0 +1,106 @@
+from __future__ import annotations
+
+from django.core.cache import cache
+from django.core.exceptions import ValidationError
+from django.test import TestCase, override_settings
+
+from .models import SensitiveWord
+from .normalization import normalize_text
+from .providers.keyword import reset_keyword_matcher_cache
+from .services import moderate_prompt
+from .versioning import get_sensitive_words_version
+
+
+@override_settings(MODERATION_CACHE_VERSION_KEY="test:moderation:sensitive_words:version")
+class SensitiveWordNormalizationTests(TestCase):
+    def setUp(self):
+        cache.clear()
+        reset_keyword_matcher_cache()
+
+    def tearDown(self):
+        reset_keyword_matcher_cache()
+        cache.clear()
+
+    def test_normalize_text_removes_separators_zero_width_and_case(self):
+        self.assertEqual(normalize_text("Ｂａｄ-\u200b Word"), "badword")
+        self.assertEqual(normalize_text("敏 感-词"), normalize_text("敏感词"))
+
+    def test_sensitive_word_saves_normalized_word_and_rejects_empty_normalized_word(self):
+        word = SensitiveWord.objects.create(word=" 敏 感-词 ", category="")
+
+        self.assertEqual(word.category, "custom")
+        self.assertEqual(word.normalized_word, "敏感词")
+
+        with self.assertRaises(ValidationError):
+            SensitiveWord.objects.create(word=" - \u200b ")
+
+    def test_word_save_bumps_shared_cache_version(self):
+        before = get_sensitive_words_version()
+
+        SensitiveWord.objects.create(word="敏感词")
+
+        self.assertNotEqual(get_sensitive_words_version(), before)
+
+
+@override_settings(
+    MODERATION_ENABLED=True,
+    MODERATION_PROVIDER="keyword",
+    MODERATION_CACHE_VERSION_KEY="test:moderation:sensitive_words:version",
+)
+class KeywordModerationTests(TestCase):
+    def setUp(self):
+        cache.clear()
+        reset_keyword_matcher_cache()
+
+    def tearDown(self):
+        reset_keyword_matcher_cache()
+        cache.clear()
+
+    def test_prompt_blocks_normalized_keyword_without_storing_prompt(self):
+        word = SensitiveWord.objects.create(word="敏感词", category="policy")
+        prompt = "请处理敏-感\u200b 词内容"
+
+        with self.assertLogs("apps.moderation.services", level="INFO") as logs:
+            outcome = moderate_prompt(prompt=prompt)
+
+        self.assertTrue(outcome.blocked)
+        self.assertEqual(outcome.labels, ("policy",))
+        self.assertEqual(outcome.keywords, ("敏感词",))
+        self.assertEqual(outcome.matched_word_ids, (word.id,))
+        self.assertNotIn(prompt, "\n".join(logs.output))
+
+    def test_non_matching_prompt_passes(self):
+        SensitiveWord.objects.create(word="敏感词")
+
+        outcome = moderate_prompt(prompt="正常业务标题")
+
+        self.assertFalse(outcome.blocked)
+
+    def test_matcher_rebuilds_after_shared_version_changes(self):
+        self.assertFalse(moderate_prompt(prompt="新增词").blocked)
+
+        SensitiveWord.objects.create(word="新增词")
+
+        self.assertTrue(moderate_prompt(prompt="这个提示包含新增词").blocked)
+
+
+@override_settings(
+    MODERATION_ENABLED=False,
+    MODERATION_PROVIDER="keyword",
+    MODERATION_CACHE_VERSION_KEY="test:moderation:sensitive_words:version",
+)
+class DisabledModerationTests(TestCase):
+    def setUp(self):
+        cache.clear()
+        reset_keyword_matcher_cache()
+
+    def tearDown(self):
+        reset_keyword_matcher_cache()
+        cache.clear()
+
+    def test_disabled_moderation_is_noop(self):
+        SensitiveWord.objects.create(word="敏感词")
+
+        outcome = moderate_prompt(prompt="敏感词")
+
+        self.assertFalse(outcome.blocked)