Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

fix: strip keyword rows in keyword moderation (#42048)

Harsh Kashyapa1bb3178926f
diff --git a/api/core/moderation/keywords/keywords.py b/api/core/moderation/keywords/keywords.py
index 339574556d9..1b12e8d27a8 100644
--- a/api/core/moderation/keywords/keywords.py
+++ b/api/core/moderation/keywords/keywords.py
@@ -42,8 +42,7 @@ class KeywordsModeration(Moderation):
if query:
inputs["query__"] = query
- # Filter out empty values
- keywords_list = [keyword for keyword in self.config["keywords"].split("\n") if keyword]
+ keywords_list = self._keyword_rows(self.config["keywords"])
flagged = self._is_violated(inputs, keywords_list)
@@ -59,8 +58,7 @@ class KeywordsModeration(Moderation):
raise ValueError("The config is not set.")
if self.config["outputs_config"]["enabled"]:
- # Filter out empty values
- keywords_list = [keyword for keyword in self.config["keywords"].split("\n") if keyword]
+ keywords_list = self._keyword_rows(self.config["keywords"])
flagged = self._is_violated({"text": text}, keywords_list)
preset_response = self.config["outputs_config"]["preset_response"]
@@ -69,6 +67,16 @@ class KeywordsModeration(Moderation):
flagged=flagged, action=ModerationAction.DIRECT_OUTPUT, preset_response=preset_response
)
+ @staticmethod
+ def _keyword_rows(keywords: str) -> list[str]:
+ """Split the keywords config into stripped, non-empty rows.
+
+ Rows keep surrounding whitespace from pasted configs (for example CRLF
+ line endings) unless stripped: a keyword ending in "\r" never matches
+ real text, and a whitespace-only row matches nearly every message.
+ """
+ return [row for row in (row.strip() for row in keywords.split("\n")) if row]
+
def _is_violated(self, inputs: dict[str, Any], keywords_list: list[str]) -> bool:
return any(self._check_keywords_in_value(keywords_list, value) for value in inputs.values())
diff --git a/api/tests/unit_tests/core/moderation/test_content_moderation.py b/api/tests/unit_tests/core/moderation/test_content_moderation.py
index 4c668ee96b7..1a2c79d3890 100644
--- a/api/tests/unit_tests/core/moderation/test_content_moderation.py
+++ b/api/tests/unit_tests/core/moderation/test_content_moderation.py
@@ -127,6 +127,24 @@ class TestKeywordsModeration:
with pytest.raises(ValueError, match="inputs_config.preset_response must be less than 100 characters"):
KeywordsModeration.validate_config("test-tenant", config)
+ def test_moderation_for_inputs_crlf_keywords_match(self, keywords_config: dict[str, Any]):
+ """Keywords pasted with Windows line endings still match."""
+ keywords_config["keywords"] = "spam\r\nfree money"
+ moderation = KeywordsModeration(app_id="app", tenant_id="tenant", config=keywords_config)
+
+ result = moderation.moderation_for_inputs({}, query="this is spam")
+
+ assert result.flagged is True
+
+ def test_moderation_whitespace_only_keyword_row_is_ignored(self, keywords_config: dict[str, Any]):
+ """A whitespace-only keyword row must not flag every message."""
+ keywords_config["keywords"] = "spam\n \n "
+ moderation = KeywordsModeration(app_id="app", tenant_id="tenant", config=keywords_config)
+
+ result = moderation.moderation_for_inputs({}, query="a perfectly normal message")
+
+ assert result.flagged is False
+
def test_moderation_for_inputs_no_violation(self, keywords_moderation: KeywordsModeration):
"""Test input moderation when no keywords are matched."""
inputs = {"user_input": "This is a clean message"}
diff --git a/api/tests/unit_tests/core/moderation/test_sensitive_word_filter.py b/api/tests/unit_tests/core/moderation/test_sensitive_word_filter.py
index 585a7cf1f71..17a5184c813 100644
--- a/api/tests/unit_tests/core/moderation/test_sensitive_word_filter.py
+++ b/api/tests/unit_tests/core/moderation/test_sensitive_word_filter.py
@@ -621,13 +621,9 @@ class TestEdgeCases:
"""Test keyword that is only spaces."""
moderation = self._create_moderation(" ")
- # Text without three consecutive spaces should not match
- result1 = moderation.moderation_for_inputs({"text": "This has spaces"})
- assert result1.flagged is False
+ result = moderation.moderation_for_inputs({"text": "This has spaces"})
- # Text with three consecutive spaces should match
- result2 = moderation.moderation_for_inputs({"text": "This has spaces"})
- assert result2.flagged is True
+ assert result.flagged is False
def test_config_not_set_error_for_inputs(self):
"""Test error when config is not set for input moderation."""