🔄 synced local 'skyvern/' with remote 'skyvern/'

skyvern/forge/sdk/artifact/storage/base.py

@@ -31,6 +31,7 @@ def _file_infos_from_download_artifacts(artifacts: list[Artifact]) -> list[FileI
                 checksum=artifact.checksum,
                 filename=filename,
                 modified_at=artifact.created_at,
+                artifact_id=artifact.artifact_id,
             )
         )
     return infos

skyvern/forge/sdk/schemas/files.py

@@ -10,3 +10,8 @@ class FileInfo(BaseModel):
     checksum: str | None = Field(None, description="SHA-256 checksum of the file")
     filename: str | None = Field(None, description="Original filename")
     modified_at: datetime | None = Field(None, description="Modified time of the file")
+    # Optional: when the FileInfo is built from a DOWNLOAD Artifact row, the
+    # row's id is carried through so persisted snapshots (e.g. block outputs)
+    # can rebuild fresh signed URLs at API-fetch time even if the snapshot's
+    # ``url`` was minted before the artifact-first read existed.
+    artifact_id: str | None = Field(None, description="Artifact row id for refresh-on-read")

skyvern/forge/sdk/schemas/tasks.py

@@ -415,6 +415,11 @@ class TaskOutput(BaseModel):
     failure_category: list[dict[str, Any]] | None = None
     downloaded_files: list[FileInfo] | None = None
     downloaded_file_urls: list[str] | None = None  # For backward compatibility
+    # Persisted artifact ids for refresh-on-read (mirrors task_screenshot_artifact_ids).
+    # When set, the workflow-run-status response rebuilds ``downloaded_files`` and
+    # ``downloaded_file_urls`` from these IDs every API fetch so a presigned URL
+    # captured at execution time never makes it to the client.
+    downloaded_file_artifact_ids: list[str] | None = None
     task_screenshots: list[str] | None = None
     workflow_screenshots: list[str] | None = None
     task_screenshot_artifact_ids: list[str] | None = None
@@ -429,6 +434,10 @@ class TaskOutput(BaseModel):
     ) -> TaskOutput:
         # For backward compatibility, extract just the URLs from FileInfo objects
         downloaded_file_urls = [file_info.url for file_info in downloaded_files] if downloaded_files else None
+        # Carry artifact ids through so the API can rebuild fresh signed URLs.
+        downloaded_file_artifact_ids = (
+            [fi.artifact_id for fi in downloaded_files if fi.artifact_id] if downloaded_files else None
+        ) or None
 
         return TaskOutput(
             task_id=task.task_id,
@@ -439,6 +448,7 @@ class TaskOutput(BaseModel):
             failure_category=task.failure_category,
             downloaded_files=downloaded_files,
             downloaded_file_urls=downloaded_file_urls,
+            downloaded_file_artifact_ids=downloaded_file_artifact_ids,
             task_screenshot_artifact_ids=task_screenshot_artifact_ids,
             workflow_screenshot_artifact_ids=workflow_screenshot_artifact_ids,
         )

skyvern/forge/sdk/workflow/service.py

@@ -54,6 +54,7 @@ from skyvern.forge import app
 from skyvern.forge.failure_classifier import classify_from_failure_reason
 from skyvern.forge.prompts import prompt_engine
 from skyvern.forge.sdk.artifact.models import Artifact, ArtifactType
+from skyvern.forge.sdk.artifact.storage.base import _file_infos_from_download_artifacts
 from skyvern.forge.sdk.cache import extraction_cache
 from skyvern.forge.sdk.cache.factory import CacheFactory
 from skyvern.forge.sdk.core import skyvern_context
@@ -514,6 +515,28 @@ class WorkflowService:
         urls = await app.ARTIFACT_MANAGER.get_share_links_with_bundle_support(artifacts)
         return [u for u in urls if u is not None]
 
+    async def _file_infos_from_download_artifact_ids(
+        self,
+        artifact_ids: list[str],
+        organization_id: str | None,
+    ) -> list[FileInfo]:
+        """Rebuild ``FileInfo`` objects from DOWNLOAD artifact IDs.
+
+        Used to refresh persisted block-output ``downloaded_files`` snapshots:
+        the URL captured at execution time may be a legacy presigned S3 URL,
+        but the artifact row has everything we need to mint a fresh signed
+        ``/v1/artifacts/{id}/content`` URL on each API fetch.
+        """
+        if not artifact_ids or not organization_id:
+            return []
+        artifacts = await app.DATABASE.artifacts.get_artifacts_by_ids(artifact_ids, organization_id)
+        if not artifacts:
+            return []
+        # Preserve the input order so block outputs render files in save order.
+        by_id = {a.artifact_id: a for a in artifacts}
+        ordered = [by_id[aid] for aid in artifact_ids if aid in by_id]
+        return _file_infos_from_download_artifacts(ordered)
+
     async def _refresh_output_screenshot_urls(
         self,
         value: Any,
@@ -529,6 +552,11 @@ class WorkflowService:
 
         For backwards compatibility with old data that stored URLs directly (now expired),
         we also check for task_id and regenerate URLs using the task_id lookup.
+
+        Also refreshes ``downloaded_files`` / ``downloaded_file_urls`` from
+        ``downloaded_file_artifact_ids`` so block outputs return short signed
+        ``/v1/artifacts/{id}/content`` URLs even when the URL captured at
+        execution time was a legacy presigned S3 URL.
         """
         if isinstance(value, dict):
             # Check if this looks like a TaskOutput with screenshot artifact IDs (new format)
@@ -548,6 +576,14 @@ class WorkflowService:
                         value["workflow_screenshot_artifact_ids"],
                         organization_id,
                     )
+                if value.get("downloaded_file_artifact_ids"):
+                    refreshed = await self._file_infos_from_download_artifact_ids(
+                        value["downloaded_file_artifact_ids"],
+                        organization_id,
+                    )
+                    if refreshed:
+                        value["downloaded_files"] = [fi.model_dump(mode="json") for fi in refreshed]
+                        value["downloaded_file_urls"] = [fi.url for fi in refreshed]
             elif has_old_format:
                 # Old format (backwards compat): regenerate URLs using task_id lookup
                 task_id = value.get("task_id")