.NET: [Bug]: Foundry Responses API based agents fail in Workflows when Foundry hosted tools are used.

### Description

When AgentWorkflowBuilder.BuildSequential chains two Foundry-backed agents, the workflow forwards the complete output[] stream from agent A as the input[] for agent B's POST /openai/v1/responses call, including Foundry server-side output items such as mcp_list_tools, web_search_call, reasoning, etc. Foundry's Responses API rejects those items on the hand-off:
```json
 {
   "error": {
     "code": "invalid_payload",
     "message": "The provided data does not match the expected schema",
     "param": "/",
     "type": "invalid_request_error",
     "details": [],
     "additionalInfo": { "request_id": "055bb009413af220ba1c4166d1254ffd" }
   }
 }
```

This breaks any AF workflow that chains two or more Foundry-backed agents where any upstream agent has any Foundry-hosted tool attached (WebSearchTool, FileSearchTool, CodeInterpreterTool, McpTool, FabricDataAgentTool,
etc.).


To verify this, I have tested calling Foundry-hosted (but unpublished) agents with no tools, published agents with no tools, and unpublished agents with locally defined tools - no errors in any of these cases.


Observed sequence:

 1. Agent a1 → POST /openai/v1/responses with tools:[WebSearchTool] → 200 OK, streams response.completed with 
output[] containing web_search_call, reasoning, and message(output_text) items.
 2. Agent a2 → POST /openai/v1/responses → 400 Bad Request invalid_payload.

Request body sent to agent a2 (redacted, preserving structure)

```json
 {
   "input": [
     { "type": "message", "role": "user",
       "content": [{ "type": "input_text", "text": "Look up insurance fraud ..." }] },
 
     { "type": "web_search_call",
       "id": "ws_...",
       "status": "completed",
       "action": { "type": "search", "query": "..." },
       "agent_reference": { "type": "agent_reference", "name": "a1", "version": "..." },
       "response_id": "resp_..." },
 
     { "type": "reasoning", "id": "rs_...",
       "agent_reference": { "type": "agent_reference", "name": "a1", "version": "..." },
       "response_id": "resp_...", "summary": [] },
 
     { "type": "message", "role": "assistant",
       "content": [{ "type": "output_text", "text": "..." }] }
   ],
   "stream": true,
   "agent_reference": { "type": "agent_reference", "name": "a2", "version": "..." }
 }
```

Response:

 400 Bad Request
 apim-request-id: 2b2712d0-37bf-43f8-9bbe-d2000bde83d6
 x-ms-region: East US

 {
   "error": {
     "code": "invalid_payload",
     "message": "The provided data does not match the expected schema",
     "param": "/",
     "type": "invalid_request_error"
   }
 }

# Potential fix:

AgentWorkflowBuilder.BuildSequential (and any other chaining builder that crosses agent boundaries) should sanitize
the output[] of each stage before feeding it as input[] to the next stage. Minimum fix:

 - Drop any non-message items: mcp_list_tools, mcp_call, mcp_approval_request, mcp_approval_response, reasoning, 
web_search_call, file_search_call, code_interpreter_call, fabric_dataagent_preview_call, image_generation_call, 
computer_tool_call, and any future server-side call/output items.
 - Strip agent_reference and response_id from any forwarded item — those are scoped to the producing response and 
invalid on a cross-agent call.
 - Alternatively, surface a first-class extension point (e.g. .WithInterAgentInputFilter(...)) so consumers can 
configure the hand-off contract.

### Code Sample

```markdown
extern alias FoundryAlias;
 using Azure.AI.Projects;
 using Azure.Identity;
 using Microsoft.Agents.AI;
 using Microsoft.Agents.AI.Workflows;
 using Microsoft.Extensions.AI;
 using OpenAI.Responses;
 using FoundryExtensions = FoundryAlias::Azure.AI.Projects.AzureAIProjectChatClientExtensions;
 using FoundryAITool    = FoundryAlias::Microsoft.Agents.AI.Foundry.FoundryAITool;
 
 var client = new AIProjectClient(new Uri(endpoint), new DefaultAzureCredential());
 
 // WebSearchTool: Foundry-hosted, no config, emits `web_search_call` output items.
 IList<AITool> tools = [FoundryAITool.FromResponseTool(new WebSearchTool())];
 
 var a1 = FoundryExtensions.AsAIAgent(client,
     model: "gpt-5.4-mini",
     instructions: "Use web search to look up something relevant, then summarize briefly.",
     name: "a1", description: null, tools: tools);
 
 var a2 = FoundryExtensions.AsAIAgent(client,
     model: "gpt-5.4-mini",
     instructions: "Summarize the prior agent's output in one sentence.",
     name: "a2", description: null, tools: tools);
 
 var workflow = AgentWorkflowBuilder.BuildSequential([a1, a2]);
 
 await using var run = await InProcessExecution.RunStreamingAsync(
     workflow, new ChatMessage(ChatRole.User, "Look up insurance fraud patterns for rear-end collisions."));
 await run.TrySendMessageAsync(new TurnToken(emitEvents: true));
 
 await foreach (var evt in run.WatchStreamAsync()) { /* observe */ }
```

### Error Messages / Stack Traces

```markdown
{
   "error": {
     "code": "invalid_payload",
     "message": "The provided data does not match the expected schema",
     "param": "/",
     "type": "invalid_request_error",
     "details": [],
     "additionalInfo": { "request_id": "055bb009413af220ba1c4166d1254ffd" }
   }
 }
ERROR: System.Reflection.TargetInvocationException: Error invoking handler for Microsoft.Agents.AI.Workflows.TurnToken
 ---> System.InvalidOperationException: Collection was modified; enumeration operation may not execute.
   at System.Collections.Generic.List`1.Enumerator.MoveNext()
   at Microsoft.Extensions.AI.ChatResponseExtensions.CoalesceWebSearchToolCallContent(IList`1 contents)
   at Microsoft.Extensions.AI.ChatResponseExtensions.CoalesceContent(IList`1 contents)
   at Microsoft.Extensions.AI.ChatResponseExtensions.FinalizeResponse(ChatResponse response)
   at Microsoft.Extensions.AI.ChatResponseExtensions.ToChatResponse(IEnumerable`1 updates)
   at Microsoft.Agents.AI.AgentResponseExtensions.ToAgentResponse(IEnumerable`1 updates)
   at Microsoft.Agents.AI.Workflows.Specialized.AIAgentHostExecutor.InvokeAgentAsync(IEnumerable`1 messages, IWorkflowContext context, Boolean emitEvents, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.Specialized.AIAgentHostExecutor.ContinueTurnAsync(List`1 messages, IWorkflowContext context, Boolean emitEvents, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.ChatProtocolExecutor.<>c__DisplayClass13_0.<<TakeTurnAsync>g__InvokeTakeTurnAsync|0>d.MoveNext()
--- End of stack trace from previous location ---
   at Microsoft.Agents.AI.Workflows.StatefulExecutor`1.InvokeWithStateAsync(Func`4 invocation, IWorkflowContext context, Boolean skipCache, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.RouteBuilder.<>c__DisplayClass12_0`1.<<AddHandler>g__WrappedHandlerAsync|0>d.MoveNext()
--- End of stack trace from previous location ---
   at Microsoft.Agents.AI.Workflows.Execution.MessageRouter.RouteMessageAsync(Object message, IWorkflowContext context, Boolean requireRoute, CancellationToken cancellationToken)
   --- End of inner exception stack trace ---
   at Microsoft.Agents.AI.Workflows.Executor.ExecuteCoreAsync(Object message, TypeId messageType, IWorkflowContext context, WorkflowTelemetryContext telemetryContext, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.InProc.InProcessRunner.DeliverMessagesAsync(String receiverId, ConcurrentQueue`1 envelopes, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.InProc.InProcessRunner.DeliverMessagesAsync(String receiverId, ConcurrentQueue`1 envelopes, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.InProc.InProcessRunner.RunSuperstepAsync(StepContext currentStep, CancellationToken cancellationToken)
   at Microsoft.Agents.AI.Workflows.InProc.InProcessRunner.Microsoft.Agents.AI.Workflows.Execution.ISuperStepRunner.RunSuperStepAsync(CancellationToken cancellationToken)
```

### Package Versions

Microsoft.Agents.AI   1.1.0, Microsoft.Agents.AI.Workflows   1.1.0, Microsoft.Agents.AI.Foundry   1.1.0 , Azure.AI.Projects   2.0.0, Azure.Identity   1.20.0 

### .NET Version

.NET 10.0

### Additional Context

I've been testing this with tools such as the Web Search tool, and WorkIQ MCP tools, all failing in the same way.


Foundry endpoint api-version=v1 (supported: 1.0, 2025-05-15-preview, 2025-11-15-preview) , Model: gpt-5.4-mini

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

.NET: [Bug]: Foundry Responses API based agents fail in Workflows when Foundry hosted tools are used. #5338

Description

Potential fix:

Code Sample

Error Messages / Stack Traces

Package Versions

.NET Version

Additional Context

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

.NET: [Bug]: Foundry Responses API based agents fail in Workflows when Foundry hosted tools are used. #5338

Description

Description

Potential fix:

Code Sample

Error Messages / Stack Traces

Package Versions

.NET Version

Additional Context

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions