feat: 데이터 비식별화(Data Anonymizer) 도구 추가

backend/api/tools.py

@@ -529,6 +529,8 @@ async def base64_decoder_handler(params: Dict[str, Any]) -> Dict[str, str]:
         "합니다",
     }
 )
+
+
 def _normalize_analysis_text(value: str) -> str:
     """Normalize user text for deterministic, multilingual rule matching."""
     if len(value) > ANALYSIS_TEXT_MAX_CHARS:
@@ -675,6 +677,49 @@ async def uuid_v4_generator_handler(params: Dict[str, Any]) -> Dict[str, str]:
     return {"uuid": str(uuid.uuid4())}
 
 
+_INTERNATIONAL_EMAIL_PATTERN = re.compile(
+    rf"(?<![\w{_EMAIL_ATOM}.-])"
+    rf"[\w{_EMAIL_ATOM}-]+(?:\.[\w{_EMAIL_ATOM}-]+)*@"
+    r"(?:[^\W_](?:(?:[^\W_]|-){0,61}[^\W_])?\.)+"
+    r"[^\W_]{2,63}(?![\w-])"
+)
+_INTERNATIONAL_PHONE_PATTERN = re.compile(
+    r"(?<!\d)(?:01[016789][ .-]?\d{3,4}[ .-]?\d{4}"
+    r"|0[1-9](?:[ .-]?\d{2}){4})(?!\d)"
+)
+_KOREAN_RESIDENT_REGISTRATION_PATTERN = re.compile(
+    r"(?<!\d)\d{6}[ -]?[1-4]\d{6}(?!\d)"
+)
+
+
+async def data_anonymizer_handler(params: Dict[str, Any]) -> Dict[str, str]:
+    """Mask bounded contact and Korean resident-registration identifiers."""
+    text = params.get("text", "")
+    if text is None:
+        text = ""
+    if len(text) > ANALYSIS_TEXT_MAX_CHARS:
+        raise ValueError(
+            f"Analysis text must not exceed {ANALYSIS_TEXT_MAX_CHARS} characters"
+        )
+    text = _EMAIL_PATTERN.sub("***@***", text)
+    text = _INTERNATIONAL_EMAIL_PATTERN.sub("***@***", text)
+    text = _PHONE_PATTERN.sub("***-****-****", text)
+    text = _INTERNATIONAL_PHONE_PATTERN.sub("***-****-****", text)
+    text = _KOREAN_RESIDENT_REGISTRATION_PATTERN.sub("******-*******", text)
+    return {"anonymized_text": text}
+
+
+registry.register(
+    ToolInfo(
+        code="data_anonymizer",
+        name="데이터 비식별화 (Data Anonymizer)",
+        description="텍스트에서 이메일 주소, 일부 한국·북미·프랑스 전화번호, 한국 주민등록번호 형식을 단순 마스킹합니다. 완전한 개인정보 비식별화를 보장하지 않습니다.",
+        category="보안",
+        parameters={"text": "string"},
+    ),
+    data_anonymizer_handler,
+)
+
 registry.register(
     ToolInfo(
         code="uuid_v4_generator",

backend/tests/test_tools_api.py

@@ -1111,6 +1111,83 @@ def test_execute_sentiment_analyzer():
     assert "불만" in data["result"]["key_emotions"]
 
 
+def test_execute_data_anonymizer():
+    with TestClient(app) as client:
+        # 정상적인 케이스
+        response = client.post(
+            "/api/tools/data_anonymizer/execute",
+            headers={"Authorization": f"Bearer {_signed_session_token()}"},
+            json={
+                "parameters": {
+                    "text": "제 이메일은 test.user-1@gmail.com 이고, 폰 번호는 010-1234-5678, 주민번호는 900101-1234567 입니다. 011-123-4567도 됩니다."
+                }
+            },
+        )
+        assert response.status_code == 200
+        data = response.json()
+        assert data["status"] == "success"
+        anonymized = data["result"]["anonymized_text"]
+        assert "***@***" in anonymized
+        assert "***-****-****" in anonymized
+        assert "******-*******" in anonymized
+        assert "test.user-1@gmail.com" not in anonymized
+        assert "010-1234-5678" not in anonymized
+        assert "900101-1234567" not in anonymized
+
+        # 빈 텍스트 케이스
+        response_empty = client.post(
+            "/api/tools/data_anonymizer/execute",
+            headers={"Authorization": f"Bearer {_signed_session_token()}"},
+            json={"parameters": {"text": ""}},
+        )
+        assert response_empty.status_code == 200
+        data_empty = response_empty.json()
+        assert data_empty["status"] == "success"
+        assert data_empty["result"]["anonymized_text"] == ""
+
+        # null 텍스트 케이스를 막는 동작은 _validate_parameters가 하지만,
+        # fallback 커버리지를 위해 직접 handler를 호출하는 비동기 테스트를 아래에 추가합니다.
+
+
+def test_execute_data_anonymizer_masks_separator_free_and_international_formats():
+    source_values = (
+        "01012345678",
+        "9001011234567",
+        "01 42 68 53 00",
+        "사용자@예시.한국",
+    )
+    with TestClient(app) as client:
+        response = client.post(
+            "/api/tools/data_anonymizer/execute",
+            headers={"Authorization": f"Bearer {_signed_session_token()}"},
+            json={"parameters": {"text": " / ".join(source_values) + "."}},
+        )
+
+    assert response.status_code == 200
+    anonymized = response.json()["result"]["anonymized_text"]
+    assert all(source_value not in anonymized for source_value in source_values)
+    assert anonymized.endswith(".")
+    assert anonymized.count("***-****-****") == 2
+    assert "******-*******" in anonymized
+    assert "***@***" in anonymized
+
+
+@pytest.mark.asyncio
+async def test_data_anonymizer_handler_none():
+    from api.tools import data_anonymizer_handler
+
+    result = await data_anonymizer_handler({"text": None})
+    assert result["anonymized_text"] == ""
+
+    result_missing = await data_anonymizer_handler({})
+    assert result_missing["anonymized_text"] == ""
+
+    with pytest.raises(ValueError, match="Analysis text must not exceed"):
+        await data_anonymizer_handler(
+            {"text": "x" * (ANALYSIS_TEXT_MAX_CHARS + 1)}
+        )
+
+
 def test_execute_grammar_checker():
     with TestClient(app) as client:
         response = client.post(

docs/doctoring/data-anonymizer-boundary.md

@@ -0,0 +1,41 @@
+# Data anonymizer boundary
+
+## Decision and observed implementation
+
+PR #1482 repair parent `034d111b6bef126929d6f0085c2fa15bbf9724be`
+stacks on PR #1555 exact head
+`03799bc157fa39a419cf6c3f77a29a2ca02cd7f4`. The handler reuses the stack's
+canonical ASCII email and selected Korean/North American phone matchers, then
+adds bounded Unicode-email, French phone, and Korean resident-registration
+patterns. Every input is subject to `ANALYSIS_TEXT_MAX_CHARS` before scanning.
+
+This tool performs deterministic format masking only. It does not measure
+re-identification risk, detect names or organizations, transform free-form
+quasi-identifiers, or certify that output is anonymous. Product copy must keep
+that limitation visible. A downstream workflow that requires release-grade
+de-identification needs a documented data model, threat model, risk metric,
+review authority, and evidence that the transformed dataset meets its intended
+use. It must not infer that assurance from this handler's successful response.
+
+Endpoint regressions cover hyphenated and separator-free Korean identifiers,
+an internationalized email address, a French phone number, punctuation
+preservation, and the input-size boundary. The values are synthetic test data;
+no real person's identifiers are committed.
+
+## Research grounding
+
+NIST SP 800-188 treats de-identification as a managed process involving data
+models, techniques, governance, and re-identification risk rather than a small
+set of textual substitutions. That distinction supports the deliberately
+narrow product claim above and rejects the earlier broad “data anonymization”
+assurance.
+
+Garfinkel, S., Guttman, B., Near, J., Dajani, A., & Singer, P. (2023).
+*De-identifying government datasets: Techniques and governance* (NIST Special
+Publication 800-188). National Institute of Standards and Technology.
+https://doi.org/10.6028/NIST.SP.800-188
+
+The official publication page was available during verification, but its
+linked PDF endpoint returned HTTP 404 on 2026-09-04. The PR therefore records
+the DOI and bounded summary instead of committing an unverified or
+redistribution-uncertain binary.