.NET: [Bug]: I see 1.13 release breaking change which was not classified as such, session not being updated when skills used

[VaclavK]

Description

Upon restructuring our agentic solution set up, we build and release in a mode that started to rely on skills. Previous build set up did not.

Upon the switch we started to see session not updated with agent results and investigation led us to

#6843

This I believe is not classified as breaking change but we could see that skill loading was awaiting approval according to session. The symptom

 {
                    "$type": "toolApprovalRequest",
                    "toolCall": {
                        "$type": "functionCall",
                        "name": "load_skill",
                        "arguments": {
                            "skillName": "flow-classifier"
                        },
                        "informationalOnly": true,
                        "callId": "call_YpVrfC7biu6hWAry8clkuDtv"
                    },
                    "requestId": "ficc_call_YpVrfC7biu6hWAry8clkuDtv"
                }

No agent result was appended to the session.

Setting

   options: new AgentSkillsProviderOptions
   {
       DisableLoadSkillApproval = true,
       DisableReadSkillResourceApproval = true,
       DisableRunSkillScriptApproval = true,
   }

produces up to date result.

What is not yet clear to me is why the agent produced the result and not explicitly setting these things only broke the session update.

Code Sample

var session = await SessionService.GetOrCreateSessionAsync(agentSessionStore, sessionAgent, request.SessionId);

  var result = await agent.RunAsync<OrchestrationEnvelope>(BuildMessage(request), session);

  await SessionService.SaveSessionAsync(sessionAgent, agentSessionStore, session, request.SessionId);

the context for this pattern - #6858

Error Messages / Stack Traces

Package Versions

Microsoft.Agents.AI, 1.13

.NET Version

.net 10

Additional Context

No response

[github-actions]

👋 Hi @VaclavK — thanks for the report!

Your toolApprovalRequest for load_skill is the expected default behavior introduced in the 1.13 release cycle — commit e3b64fd (PR #6729) is explicitly labeled [BREAKING] and wraps all skill tools with ApprovalRequiredAIFunction by default as a security measure. The options you found (DisableLoadSkillApproval = true, etc.) are the documented opt-out mechanism. Alternatively, you can use AgentSkillsProvider.AllToolsAutoApprovalRule or AgentSkillsProvider.ReadOnlyToolsAutoAprovalRule as auto-approval rules on your agent, which avoids disabling approval entirely while still allowing automatic execution.

If you still believe there is a framework issue here, please reply with the exact import/class or method you are using, the version, and a minimal snippet that shows the failure. 🙏

[VaclavK]

AllToolsAutoApprovalRule

this is a good pointer

we had already - prior to decorating these additional flags this in place

   return new ChatClientAgent(chatClient, options)
       .AsBuilder()
       .UseToolApproval(new ToolApprovalAgentOptions { AutoApprovalRules = [autoApprove] })
       .Build();

with

Func<FunctionCallContent, ValueTask<bool>> autoApprove = (functionCall) =>
{
    // Auto-approve all function calls for testing purposes
    return ValueTask.FromResult(true);
};

but this clearly had no bearing on skill tool only non-skill tooling

[VaclavK]

can I ask what is the recommended strategy to track breaking changes..?

continuous scanning of this repo for the titles of merges and analysing content per each to understand impact?