fix: preserve single files for fallback extraction paths

src/openai/_utils/_utils.py

@@ -114,7 +114,7 @@ def _extract_items(
         try:
             # Remove the field if there are no more dict keys in the path,
             # only "<array>" traversal markers or end.
-            if all(p == "<array>" for p in path[index:]):
+            if all(p == "<array>" for p in path[index:]) and (index == len(path) or is_list(obj[key])):
                 item = obj.pop(key)
             else:
                 item = obj[key]

tests/test_extract_files.py

@@ -41,6 +41,16 @@ def test_top_level_file_array() -> None:
     assert query == {"title": "hello"}
 
 
+@pytest.mark.parametrize("file", [b"zip contents", ("skill.zip", b"zip contents")])
+def test_single_file_fallback_after_array_path(file: FileTypes) -> None:
+    query = {"files": file, "title": "hello"}
+    assert extract_files(query, paths=[["files", "<array>"]]) == []
+    assert query == {"files": file, "title": "hello"}
+
+    assert extract_files(query, paths=[["files", "<array>"], ["files"]]) == [("files", file)]
+    assert query == {"title": "hello"}
+
+
 @pytest.mark.parametrize(
     "query,paths,expected",
     [

tests/test_files.py

@@ -1,14 +1,56 @@
+import io
+import zipfile
 from pathlib import Path
 
 import anyio
+import httpx2
 import pytest
 
+from openai import OpenAI, AsyncOpenAI
 from openai._files import to_httpx_files, deepcopy_with_paths, async_to_httpx_files
 from openai._utils import extract_files
 
 readme_path = Path(__file__).parent.parent.joinpath("README.md")
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("is_async", [False, True])
+@pytest.mark.parametrize("as_list", [False, True])
+async def test_skills_upload_zip(is_async: bool, as_list: bool) -> None:
+    buffer = io.BytesIO()
+    with zipfile.ZipFile(buffer, "w") as archive:
+        archive.writestr("SKILL.md", "---\nname: example\ndescription: Test skill\n---\n# Example\n")
+    content = buffer.getvalue()
+    file = ("skill.zip", content)
+    files = [file] if as_list else file
+    transport = httpx2.MockTransport(lambda _request: httpx2.Response(200, json={}))
+    client = (
+        AsyncOpenAI(
+            api_key="fake-key", base_url="https://example.test", http_client=httpx2.AsyncClient(transport=transport)
+        )
+        if is_async
+        else OpenAI(api_key="fake-key", base_url="https://example.test", http_client=httpx2.Client(transport=transport))
+    )
+    try:
+        response = (
+            await client.skills.with_raw_response.create(files=files)
+            if isinstance(client, AsyncOpenAI)
+            else client.skills.with_raw_response.create(files=files)
+        )
+        request = response.http_request
+        assert request.url.path == "/skills"
+        assert request.headers["content-type"].startswith("multipart/form-data;")
+        expected_field = b'name="files[]"' if as_list else b'name="files"'
+        assert expected_field in request.content
+        assert b'filename="skill.zip"' in request.content
+        assert request.content.count(content) == 1
+    finally:
+        if isinstance(client, AsyncOpenAI):
+            await client.close()
+        else:
+            client.close()
+
+
 def test_pathlib_includes_file_name() -> None:
     result = to_httpx_files({"file": readme_path})
     assert result == {"file": ("README.md", readme_path.read_bytes())}