Responses API requests fail with "Unsupported parameter: 'reasoning_effort'" when wireApi is 'responses'

[pmella]

Environment

• @github/copilot-sdk: 0.1.32
• @github/copilot (CLI): 1.0.6

Description

When ProviderConfig.wireApi is set to 'responses', the CLI still sends reasoning_effort as a top-level request parameter. The OpenAI Responses API rejects this with:

400 {"error":{"message":"Unsupported parameter: 'reasoning_effort'. In the Responses API, this parameter has moved to 'reasoning.effort'."}}

The parameter name change is documented at: https://platform.openai.com/docs/api-reference/responses/create

Root cause

In getCompletionOptions(), the CLI always returns { reasoning_effort: o, ... } regardless of wire API format. When using the Responses API, this should be { reasoning: { effort: o } }.

Impact

All GPT model requests fail when wireApi: 'responses' is enabled. Affects both GPT-52-reasoning and GPT-54-reasoning.

Repro

1. Set provider.wireApi = 'responses' in session config
2. Send any prompt to a GPT reasoning model
3. First LLM call fails with 400

Expected behavior

The CLI should adapt the parameter format based on the wire API — reasoning_effort for Chat Completions, reasoning.effort for Responses API.

Current workaround

Disabled Responses API via feature flag (ExcelAgent.EnableResponsesApi defaulting to false).

[SteveSandersonMS]

@pmella Could you clarify the repro slightly more?

Set provider.wireApi = 'responses' in session config

Presumably it's also necessary to set up other provider params, is it? Are you using BYOK with an OpenAI endpoint?

[SteveSandersonMS]

Maybe this was already fixed. I was unable to reproduce the issue. I created a session using BYOK and Azure OpenAI like this:

    const session = await client.createSession({
        model: "gpt-5.4",
        reasoningEffort: "medium",
        onPermissionRequest: approveAll,
        provider: {
            baseUrl: "https://MY-DEPLOYMENT.openai.azure.com/openai/v1",
            apiKey: "MY-API-KEY",
            type: "openai",
            wireApi: "responses",
        },
    });

... and it correctly responds without errors.

If you can still repro the issue on the current version of the SDK, can you post standalone repro code? Thanks!