Skip to content

Automate browser based workflows with AI

fix(downloads): recover missing file extension by sniffing content (#7470)

Shuchang Zhengee411c98c06e
diff --git a/skyvern/forge/agent.py b/skyvern/forge/agent.py
index 66402d8dc..fd14bd8e1 100644
--- a/skyvern/forge/agent.py
+++ b/skyvern/forge/agent.py
@@ -78,6 +78,7 @@ from skyvern.forge.sdk.api.files import (
get_path_for_workflow_download_directory,
list_downloading_files_in_directory,
list_files_in_directory,
+ recover_download_extension,
rename_file,
resolve_run_download_id,
wait_for_download_finished,
@@ -537,6 +538,19 @@ class ForgeAgent:
)
continue
+ if not file_extension:
+ file_extension = recover_download_extension(
+ os.path.join(workflow_download_directory, local_file_name), download_suffix
+ )
+ if file_extension:
+ LOG.info(
+ "Recovered missing download file extension from file content",
+ file=local_file_name,
+ extension=file_extension,
+ task_id=task.task_id,
+ workflow_run_id=task.workflow_run_id,
+ )
+
if download_suffix:
# local_file_name is a bare basename for session (s3/gs) files but an absolute path for
# run-dir files; compare on basenames so a file already named by download_suffix is not
diff --git a/skyvern/forge/sdk/api/files.py b/skyvern/forge/sdk/api/files.py
index e090b5d72..abf0b22fc 100644
--- a/skyvern/forge/sdk/api/files.py
+++ b/skyvern/forge/sdk/api/files.py
@@ -12,6 +12,7 @@ from typing import TYPE_CHECKING
from urllib.parse import parse_qsl, unquote, urlparse
import aiohttp
+import filetype
import structlog
from multidict import CIMultiDictProxy
from yarl import URL
@@ -576,6 +577,26 @@ def sanitize_filename(filename: str) -> str:
return "".join(c for c in filename if c.isalnum() or c in ["-", "_", ".", "%", " "])
+def guess_extension_from_file(file_path: str | Path) -> str:
+ """Infer a file's extension (with leading dot) from its magic bytes, or "" if unreadable/unknown."""
+ try:
+ kind = filetype.guess(str(file_path))
+ except OSError:
+ return ""
+ return f".{kind.extension}" if kind else ""
+
+
+def recover_download_extension(file_path: str | Path, download_suffix: str | None = None) -> str:
+ """Extension to append to a downloaded file that has none, sniffed from its content.
+
+ Returns "" when ``download_suffix`` already carries its own extension, so the final
+ ``download_suffix + extension`` name is not doubled (e.g. invoice.pdf + .pdf).
+ """
+ if download_suffix and Path(download_suffix).suffix:
+ return ""
+ return guess_extension_from_file(file_path)
+
+
def rename_file(file_path: str, new_file_name: str) -> str:
try:
new_file_name = sanitize_filename(new_file_name)
diff --git a/skyvern/services/script_service.py b/skyvern/services/script_service.py
index 5437d24b0..8e3e70c48 100644
--- a/skyvern/services/script_service.py
+++ b/skyvern/services/script_service.py
@@ -40,6 +40,7 @@ from skyvern.forge.sdk.api.files import (
check_downloading_files_and_wait_for_download_to_complete,
get_path_for_workflow_download_directory,
list_files_in_directory,
+ recover_download_extension,
rename_file,
resolve_run_download_id,
)
@@ -2322,6 +2323,14 @@ async def download(
# Skip incomplete downloads
if file_extension == BROWSER_DOWNLOADING_SUFFIX:
continue
+ if not file_extension:
+ file_extension = recover_download_extension(file_path, download_suffix)
+ if file_extension:
+ LOG.info(
+ "Recovered missing download file extension from file content",
+ file=file_path,
+ extension=file_extension,
+ )
local_basename = Path(file_path).name
existing_names = {
Path(f).name
diff --git a/tests/unit/test_download_extension_sniffing.py b/tests/unit/test_download_extension_sniffing.py
new file mode 100644
index 000000000..30e31bf6d
--- /dev/null
+++ b/tests/unit/test_download_extension_sniffing.py
@@ -0,0 +1,50 @@
+from skyvern.forge.sdk.api.files import guess_extension_from_file, recover_download_extension
+
+_PDF_BYTES = b"%PDF-1.4\n%\xd3\xeb\xe9\xe1\n1 0 obj\n<</Title (statement)>>\nrest of the document"
+
+
+def test_guess_extension_from_file_detects_pdf_without_extension(tmp_path) -> None:
+ extensionless = tmp_path / "download-20260617114038544470-ABCD"
+ extensionless.write_bytes(_PDF_BYTES)
+
+ assert guess_extension_from_file(extensionless) == ".pdf"
+
+
+def test_guess_extension_from_file_returns_empty_for_unknown_content(tmp_path) -> None:
+ unknown = tmp_path / "mystery"
+ unknown.write_bytes(b"this header matches nothing")
+
+ assert guess_extension_from_file(unknown) == ""
+
+
+def test_guess_extension_from_file_returns_empty_for_missing_file(tmp_path) -> None:
+ assert guess_extension_from_file(tmp_path / "does-not-exist") == ""
+
+
+def test_recover_download_extension_sniffs_when_no_download_suffix(tmp_path) -> None:
+ downloaded = tmp_path / "download-xyz"
+ downloaded.write_bytes(_PDF_BYTES)
+
+ assert recover_download_extension(downloaded, download_suffix=None) == ".pdf"
+
+
+def test_recover_download_extension_sniffs_when_download_suffix_lacks_extension(tmp_path) -> None:
+ downloaded = tmp_path / "download-xyz"
+ downloaded.write_bytes(_PDF_BYTES)
+
+ assert recover_download_extension(downloaded, download_suffix="invoice") == ".pdf"
+
+
+def test_recover_download_extension_skips_when_download_suffix_already_has_extension(tmp_path) -> None:
+ # Guard against invoice.pdf + .pdf -> invoice.pdf.pdf
+ downloaded = tmp_path / "download-xyz"
+ downloaded.write_bytes(_PDF_BYTES)
+
+ assert recover_download_extension(downloaded, download_suffix="invoice.pdf") == ""
+
+
+def test_recover_download_extension_empty_for_unrecognized_content(tmp_path) -> None:
+ downloaded = tmp_path / "download-xyz"
+ downloaded.write_bytes(b"not a recognizable file")
+
+ assert recover_download_extension(downloaded, download_suffix=None) == ""