Repository navigation
Python: HandoffBuilder: Generated file cannot be retrieved #2718
Description
My scenario is that I receive a large response from an external tool that I want to transform into a file. Then, this file should be processed by other agents, who will create additional files from it. From my test this happens, as the 2 file agents pass this file id and generate new files. But at the end of the workflow I cannot download the file.
...
file_gen_agent = azure_ai_client.create_agent(
name="FileGeneratorAgent",
instructions=(
"You are an excellent text file generator. "
"Call the Code Interpreter tool to generate a txt file based on the user's request. "
),
tools=HostedCodeInterpreterTool(),
middleware=[
override_middleware,
comomon_session_data.inject_context_middleware,
],
store=True,
)
...
workflow_agent = (
HandoffBuilder(
name="SampleWorkflowAgent",
participants=[coordinator_agent, file_gen_agent, pdf_file_gen_agent],
)
.set_coordinator(coordinator_agent)
.add_handoff(coordinator_agent, file_gen_agent)
.auto_register_handoff_tools(enabled=True)
.add_handoff(file_gen_agent, pdf_file_gen_agent)
.with_checkpointing(storage)
.build()
events = workflow_agent.run_stream(prompt)
async for event in events:
....
testFile = await self.agents_client.files.get(file_id=file_id)When running, I get this response
{
"type": "agent_run_response",
"messages": [
{
"type": "chat_message",
"role": {
"type": "role",
"value": "assistant"
},
"contents": [
{
"type": "text",
"text": "from fpdf import FPDF\n\n# Create instance of FPDF class\npdf = FPDF()\npdf.add_page()\npdf.set_font(\"Arial\", size=12)\n\n# Add sample content\nsample_content = \"\"\"\nSample PDF Document\n\nThis is a sample PDF file generated with Python.\nIt demonstrates how to create PDF files programmatically.\n\n- Item 1\n- Item 2\n- Item 3\n\nThank you for using this service!\n\"\"\"\n\nfor line in sample_content.split('\\n'):\n pdf.cell(200, 10, txt=line, ln=True)\n\n# Save PDF file\npdf_file_path = \"/mnt/data/final.pdf\"\npdf.output(pdf_file_path)\n\npdf_file_path"
},
{
"type": "text",
"annotations": [
{
"type": "citation",
"annotated_regions": [
{
"type": "text_span",
"start_index": 97,
"end_index": 124
}
],
"url": "final.pdf",
"file_id": "cfile_6937d9e89be8819090002bfae3a00fcf",
"container_id": "cntr_6937d9df20508190a73b97efbacff0bc01c8c675e5b4f541"
}
],
"text": "Your PDF file named 'final.pdf' with sample content has been generated.\n\n[Download the PDF file](sandbox:/mnt/data/final.pdf)"
}
],
"author_name": "PDFFileGeneratorAgent",
"additional_properties": {}
}
],
"response_id": "resp_00e0de3e0d80a057006937d9ddbe90819489c33df5e2180656",
"created_at": "2025-12-09T10:12:13.000000Z",
"usage_details": {
"type": "usage_details",
"input_token_count": 553,
"output_token_count": 176,
"total_token_count": 729
},
"additional_properties": {}
}However if I search for the file id cfile_6937d9e89be8819090002bfae3a00fcf I don't get it, also if I list the files it isn't there and also if I check Azure Foundry (new) I don't see the file.
Metadata
Metadata
Assignees
Labels
Type
Projects
- Status
Activity
moonbox3 commentedon Dec 10, 2025
Thanks for filing, Tamás (Tamas) (@furoTmark-ll). I tracked this issue down to the underlying AzureAIAgentClient / AzureAIClient. The clients are not properly extracting the file references from MessageDeltaChunk annotations. I've got a PR that fixes this: #2741. Note: the PR also includes a sample showing a handoff with these underlying agents and properly retrieving the file ids from a hosted code interpreter tool.