Skip to content

Automate browser based workflows with AI

Fix CDP download interceptor missing XHR-triggered downloads (#4934)

LawyZheng44d3635cef01
diff --git a/skyvern/webeye/cdp_download_interceptor.py b/skyvern/webeye/cdp_download_interceptor.py
index 0a1e9b9fe..40f06395a 100644
--- a/skyvern/webeye/cdp_download_interceptor.py
+++ b/skyvern/webeye/cdp_download_interceptor.py
@@ -36,13 +36,9 @@ MAX_FILE_SIZE_BYTES = 100 * 1024 * 1024 # 100 MB
# Resource types that should NEVER be treated as downloads.
# Sub-resources (Font, Stylesheet, etc.) are loaded by the page, not user-initiated.
-# XHR/Fetch are programmatic JS API calls (e.g. Google APIs return Content-Disposition:
-# attachment on JSON responses, which would cause false positives).
# Real user downloads come through as "Document" (link click / navigation).
NON_DOWNLOAD_RESOURCE_TYPES = frozenset(
{
- "XHR",
- "Fetch",
"Font",
"Stylesheet",
"Script",
@@ -57,6 +53,14 @@ NON_DOWNLOAD_RESOURCE_TYPES = frozenset(
}
)
+# XHR/Fetch are programmatic JS API calls that sometimes carry Content-Disposition:
+# attachment (e.g. Google APIs on JSON responses). We don't fully block them —
+# instead, we only allow them through if there's an explicit attachment header,
+# and rely on NON_DOWNLOAD_CONTENT_TYPES to filter out API false-positives.
+# Without an explicit attachment header, we skip XHR/Fetch to avoid MIME-only
+# false positives.
+XHR_FETCH_RESOURCE_TYPES = frozenset({"XHR", "Fetch"})
+
# Content types that are clearly API / data responses, never user-facing downloads,
# even if the server includes Content-Disposition: attachment.
NON_DOWNLOAD_CONTENT_TYPES = frozenset(
@@ -117,10 +121,13 @@ def is_download_response(headers: dict[str, str], status_code: int, resource_typ
Determine if a response is a file download.
Checks:
- 0. Skip sub-resource types (Font, Stylesheet, Script, Image, etc.)
- 1. Skip API content types (application/json, etc.)
- 2. Content-Disposition contains "attachment"
- 3. Content-Type is a known download MIME type
+ 0. Skip error responses (status >= 400)
+ 1. Skip sub-resource types (Font, Stylesheet, Script, Image, etc.)
+ 2. Skip API content types (application/json, etc.)
+ 3. For XHR/Fetch: require BOTH attachment header AND download MIME type
+ (prevents false positives like Google's text/plain + attachment XHR responses)
+ 4. Content-Disposition contains "attachment"
+ 5. Content-Type is a known download MIME type
"""
if status_code >= 400:
return False
@@ -134,10 +141,18 @@ def is_download_response(headers: dict[str, str], status_code: int, resource_typ
if content_type in NON_DOWNLOAD_CONTENT_TYPES:
return False
- if "attachment" in content_disposition.lower():
+ is_attachment = "attachment" in content_disposition.lower()
+ is_download_mime = content_type in DOWNLOAD_MIME_TYPES
+
+ # XHR/Fetch require both signals to avoid false positives
+ # (e.g. Google async requests: text/plain + attachment; filename="f.txt")
+ if resource_type in XHR_FETCH_RESOURCE_TYPES:
+ return is_attachment and is_download_mime
+
+ if is_attachment:
return True
- if content_type in DOWNLOAD_MIME_TYPES:
+ if is_download_mime:
return True
return False
diff --git a/tests/unit/test_cdp_download_interceptor.py b/tests/unit/test_cdp_download_interceptor.py
index 653573c3f..acfa740b4 100644
--- a/tests/unit/test_cdp_download_interceptor.py
+++ b/tests/unit/test_cdp_download_interceptor.py
@@ -65,15 +65,63 @@ class TestIsDownloadResponse:
def test_empty_headers_not_download(self) -> None:
assert is_download_response({}, 200) is False
- # Resource type filtering
- def test_xhr_resource_type_not_download(self) -> None:
+ # Resource type filtering — XHR/Fetch require BOTH attachment AND download MIME
+ def test_xhr_attachment_and_download_mime_is_download(self) -> None:
+ """XHR with attachment + download MIME type should be treated as download."""
headers = {"content-disposition": "attachment", "content-type": "application/octet-stream"}
+ assert is_download_response(headers, 200, resource_type="XHR") is True
+
+ def test_fetch_attachment_and_download_mime_is_download(self) -> None:
+ """Fetch with attachment + download MIME type should be treated as download."""
+ headers = {"content-disposition": "attachment", "content-type": "application/pdf"}
+ assert is_download_response(headers, 200, resource_type="Fetch") is True
+
+ def test_xhr_attachment_pdf_is_download(self) -> None:
+ """Real-world case: XHR download with attachment header and PDF content-type."""
+ headers = {
+ "content-disposition": "attachment; filename=Invoice_12345.pdf; filename*=UTF-8''Invoice_12345.pdf",
+ "content-type": "application/pdf",
+ }
+ assert is_download_response(headers, 200, resource_type="XHR") is True
+
+ # XHR/Fetch with attachment but non-download MIME should NOT be download
+ def test_xhr_attachment_text_plain_not_download(self) -> None:
+ """Google-style XHR: text/plain + attachment should NOT be treated as download."""
+ headers = {"content-disposition": 'attachment; filename="f.txt"', "content-type": "text/plain; charset=UTF-8"}
+ assert is_download_response(headers, 200, resource_type="XHR") is False
+
+ def test_xhr_attachment_text_html_not_download(self) -> None:
+ """XHR with text/html + attachment should NOT be treated as download."""
+ headers = {"content-disposition": "attachment", "content-type": "text/html"}
assert is_download_response(headers, 200, resource_type="XHR") is False
- def test_fetch_resource_type_not_download(self) -> None:
+ def test_fetch_attachment_only_not_download(self) -> None:
+ """Fetch with attachment but no download MIME type should NOT be download."""
headers = {"content-disposition": "attachment"}
assert is_download_response(headers, 200, resource_type="Fetch") is False
+ def test_xhr_attachment_csv_not_download(self) -> None:
+ """Known limitation: CSV via XHR is not detected because text/csv is not in DOWNLOAD_MIME_TYPES."""
+ headers = {"content-disposition": "attachment", "content-type": "text/csv"}
+ assert is_download_response(headers, 200, resource_type="XHR") is False
+
+ # XHR/Fetch without attachment header should NOT be download (MIME-only false positive)
+ def test_xhr_mime_only_not_download(self) -> None:
+ """XHR with download MIME type but no attachment header should NOT be treated as download."""
+ headers = {"content-type": "application/pdf"}
+ assert is_download_response(headers, 200, resource_type="XHR") is False
+
+ def test_fetch_mime_only_not_download(self) -> None:
+ """Fetch with download MIME type but no attachment header should NOT be treated as download."""
+ headers = {"content-type": "application/octet-stream"}
+ assert is_download_response(headers, 200, resource_type="Fetch") is False
+
+ # XHR/Fetch with attachment but API content-type should still be filtered
+ def test_xhr_json_attachment_not_download(self) -> None:
+ """XHR with JSON content-type and attachment header should NOT be download."""
+ headers = {"content-disposition": "attachment", "content-type": "application/json"}
+ assert is_download_response(headers, 200, resource_type="XHR") is False
+
def test_font_resource_type_not_download(self) -> None:
headers = {"content-type": "application/octet-stream"}
assert is_download_response(headers, 200, resource_type="Font") is False