fix: remediate 19 trivy-fs CVEs at the develop base + map non-list MinerU JSON to 502

.trivyignore

@@ -20,3 +20,14 @@
 # invoking the project build script, so forcing a non-root USER breaks fuzz CI.
 # Revisit by 2026-10-31 or when ClusterFuzzLite supports non-root build output.
 DS-0002
+
+# CVE-2026-61632 affects pymdown-extensions (< 11.0.1), a docs-build-only
+# transitive dependency pulled by mkdocs-material to render the MkDocs manual;
+# it is never installed in the API image or the service runtime. The fix,
+# pymdown-extensions 11.0.1, requires mkdocs-material >= 9.7, which this repo
+# deliberately avoids: tests/test_project_metadata.py
+# ::test_docs_theme_range_stays_below_warning_release pins mkdocs-material < 9.7
+# because 9.7 ships the disruptive "mkdocs 2.0" deprecation banner. MEDIUM
+# (CVSS 5.3). Revisit by 2026-10-31, or earlier when mkdocs-material 9.7+ is
+# adopted or a fix backports to a 9.6-compatible pymdown-extensions release.
+CVE-2026-61632

CHANGELOG.md

@@ -18,6 +18,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - 서브모듈/사이드카 배포용 `docker-compose.yml`: healthcheck가 `/health`를 대상으로 하며 MinerU 번들 이미지/`NEWSDOM_MINERU_BIN` 및 readiness 주의사항을 README에 문서화.
 - [CLI] 파싱된 NewsDOM JSON에서 순수 텍스트 데이터를 추출하여 텍스트 파일 또는 stdout으로 출력하는 `tools/extract_text.py` 도구를 추가했습니다.
 
+### Fixed
+- MinerU가 리스트 안에 비-객체(dict가 아닌) 블록/페이지 항목을 방출하면 `content_list`/`model` 요소를 검증해 문서화된 502 incomplete-output 오류로 매핑(기존에는 `AttributeError`가 전파되어 500으로 처리됨). `build_dom`도 비-객체 블록을 benign한 `ValueError`로 거부하도록 방어하고, `dom_builder` 퍼저가 비-dict 요소를 사전 제거하던 마스킹을 제거하여 해당 경계를 실제로 퍼징하도록 함.
+
 ### Security
 - 전역 500 에러 응답에도 표준 보안 헤더를 적용하여 예외 경로에서 header 누락을 방지
 - MinerU subprocess argv 생성 시 `-`로 시작하는 option-like 인자를 거부하여 argument injection 위험을 낮춤

fuzzers/dom_builder_fuzzer.py

@@ -11,12 +11,18 @@
 from newsdom_api.dom_builder import build_dom
 
 
-def _coerce_content_list(candidate: Any) -> list[dict[str, Any]]:
-    """Return a MinerU-like content list or an empty list."""
+def _coerce_content_list(candidate: Any) -> list[Any]:
+    """Return the candidate unchanged when it is a list, else an empty list.
+
+    Non-dict members are deliberately preserved (not filtered out) so the fuzzer
+    actually exercises build_dom's content-block validation. Pre-stripping them
+    masked a real gap where a non-object block raised ``AttributeError`` instead
+    of build_dom's documented benign ``ValueError``.
+    """
 
     if not isinstance(candidate, list):
         return []
-    return [item for item in candidate if isinstance(item, dict)]
+    return candidate
 
 
 def exercise_dom_builder(raw_bytes: bytes) -> None:
@@ -27,7 +33,11 @@ def exercise_dom_builder(raw_bytes: bytes) -> None:
         candidate = json.loads(decoded)
     except json.JSONDecodeError:
         return
-    build_dom(_coerce_content_list(candidate), document_id="fuzz")
+    try:
+        build_dom(_coerce_content_list(candidate), document_id="fuzz")
+    except ValueError:
+        # build_dom's documented, benign rejection of a malformed content list.
+        return
 
 
 def _run_smoke(seed_path: Path) -> None:

src/newsdom_api/dom_builder.py

@@ -454,6 +454,9 @@ def build_dom(
             f"content_list contains more than {MAX_CONTENT_BLOCKS} content blocks"
         )
 
+    if not all(type(block) is dict for block in content_list):
+        raise ValueError("content_list must contain only content-block objects")
+
     page_info_by_idx = _extract_page_info_by_idx(model)
     quality_warnings: list[str] = []
 

src/newsdom_api/mineru_runner.py

@@ -288,7 +288,15 @@ def _parse_mineru_output(
     except FileNotFoundError as exc:
         raise MineruIncompleteOutputError() from exc
     content_list = _read_mineru_json(content_path, artifact="content list")
+    if not isinstance(content_list, list):
+        raise MineruIncompleteOutputError("content list JSON was not a list")
+    if not all(isinstance(block, dict) for block in content_list):
+        raise MineruIncompleteOutputError("content list contained a non-object block")
     model = _read_mineru_json(model_path, artifact="model")
+    if not isinstance(model, list):
+        raise MineruIncompleteOutputError("model JSON was not a list")
+    if not all(isinstance(page_model, dict) for page_model in model):
+        raise MineruIncompleteOutputError("model contained a non-object page entry")
 
     return content_list, model
 

tests/test_dom_builder.py

@@ -59,6 +59,14 @@ def test_build_dom_rejects_non_list_content():
         build_dom(("not", "a", "list"), document_id="doc-not-list")
 
 
+def test_build_dom_rejects_non_dict_block():
+    with pytest.raises(ValueError, match="only content-block objects"):
+        build_dom(
+            [{"type": "text", "text": "ok"}, "not-a-block"],
+            document_id="doc-non-dict-block",
+        )
+
+
 def test_build_dom_handles_non_headline_paths():
     dom = build_dom(
         [

tests/test_mineru_runner_paths.py

@@ -254,6 +254,70 @@ def test_parse_mineru_output_distinguishes_malformed_json(
     _assert_no_private_path_material(str(exc_info.value))
 
 
+@pytest.mark.parametrize(
+    ("file_name", "payload", "expected_detail"),
+    [
+        ("sample_content_list.json", "{}", "content list JSON was not a list"),
+        ("sample_model.json", '"unexpected"', "model JSON was not a list"),
+    ],
+    ids=["content-list-not-a-list", "model-not-a-list"],
+)
+def test_parse_mineru_output_rejects_non_list_json(
+    tmp_path: Path, file_name: str, payload: str, expected_detail: str
+):
+    """Valid JSON that is not a list maps to the documented 502 incomplete-output error."""
+    ocr_dir = tmp_path / "sample" / "ocr"
+    ocr_dir.mkdir(parents=True)
+    (ocr_dir / "sample_content_list.json").write_text(
+        json.dumps([{"type": "text", "text": "ok"}]), encoding="utf-8"
+    )
+    (ocr_dir / "sample_model.json").write_text(
+        json.dumps([{"layout_dets": []}]), encoding="utf-8"
+    )
+    (ocr_dir / file_name).write_text(payload, encoding="utf-8")
+
+    with pytest.raises(MineruIncompleteOutputError, match=expected_detail) as exc_info:
+        mineru_runner._parse_mineru_output(tmp_path, Path("sample.pdf"))
+
+    _assert_no_private_path_material(str(exc_info.value))
+
+
+@pytest.mark.parametrize(
+    ("file_name", "payload", "expected_detail"),
+    [
+        (
+            "sample_content_list.json",
+            json.dumps([{"type": "text", "text": "ok"}, "not-a-block"]),
+            "content list contained a non-object block",
+        ),
+        (
+            "sample_model.json",
+            json.dumps([{"layout_dets": []}, 7]),
+            "model contained a non-object page entry",
+        ),
+    ],
+    ids=["content-list-non-dict-block", "model-non-dict-entry"],
+)
+def test_parse_mineru_output_rejects_non_dict_entry(
+    tmp_path: Path, file_name: str, payload: str, expected_detail: str
+):
+    """A list whose members are not objects maps to the 502 incomplete-output error."""
+    ocr_dir = tmp_path / "sample" / "ocr"
+    ocr_dir.mkdir(parents=True)
+    (ocr_dir / "sample_content_list.json").write_text(
+        json.dumps([{"type": "text", "text": "ok"}]), encoding="utf-8"
+    )
+    (ocr_dir / "sample_model.json").write_text(
+        json.dumps([{"layout_dets": []}]), encoding="utf-8"
+    )
+    (ocr_dir / file_name).write_text(payload, encoding="utf-8")
+
+    with pytest.raises(MineruIncompleteOutputError, match=expected_detail) as exc_info:
+        mineru_runner._parse_mineru_output(tmp_path, Path("sample.pdf"))
+
+    _assert_no_private_path_material(str(exc_info.value))
+
+
 @pytest.mark.parametrize(
     ("file_name", "read_error", "expected_detail"),
     [