Surface the Copilot usage model in response metadata

[PratikDhanave (PratikDhanave)]

The Copilot usage handler (assistantUsageUpdate) never read AssistantUsageData.Model, so a caller could not tell which model actually served the request (relevant when the Copilot backend auto-selects or falls back to a different model).

The Python client surfaces it as additional_properties={"model": model} on both run paths.

Change

• In assistantUsageUpdate, set AdditionalProperties["model"] from data.Model when non-empty. AssistantUsageData.Model exists in the pinned copilot SDK; ResponseUpdate.AdditionalProperties already propagates to the collected response. No new public API.

Test

• TestConvertToAgentResponseUpdate_UsageEvent_SurfacesModel: a usage event with model:"gpt-5" now surfaces AdditionalProperties["model"] == "gpt-5". Fails before (empty), passes after.

[Copilot]

Copilot review overview

🟢 Approval recommended

The change is small, targeted, and includes a regression test covering the new behavior.

Review effort: Lite
Findings: 1

Open (1)

• Avoid overwriting existing AdditionalProperties when setting model · New

What changed in this PR

This PR ensures the Copilot provider surfaces the actual model that served a request by copying AssistantUsageData.Model into response metadata (ResponseUpdate.AdditionalProperties["model"]), enabling callers to detect backend auto-selection or fallback behavior without introducing a new public API.

Changes:

• Populate ResponseUpdate.AdditionalProperties["model"] from AssistantUsageData.Model when non-empty.
• Add a unit test asserting the model value is present on the collected response’s AdditionalProperties.

File	Description
provider/​copilotprovider/​copilot.go	Adds propagation of the serving model into ResponseUpdate.AdditionalProperties.
provider/​copilotprovider/​copilot_test.go	Adds a regression test that verifies the model is surfaced via response metadata.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[github-actions]

Scope: user-visible behavior

Changed Go contract: assistantUsageUpdate in provider/copilotprovider/copilot.go now sets ResponseUpdate.AdditionalProperties["model"] from AssistantUsageData.Model (when non-empty), so it propagates onto the collected AgentResponse. No new exported symbols — an existing map is populated.

Upstream evidence reviewed:

• Python: python/packages/github_copilot/agent_framework_github_copilot/_agent.py — both run paths build the usage update with additional_properties={"model": model} if model else None (around lines 973/1023 for the streaming path and 1120/1127 for the non-streaming path), sourced from event.data.model. This is an exact match for naming (model key), conditional gating (only set when non-empty/non-None), and source field.
• .NET: dotnet/src/Microsoft.Agents.AI.GitHub.Copilot/GitHubCopilotAgent.cs, ConvertToAgentResponseUpdate(AssistantUsageEvent) / GetAdditionalCounts — currently only surfaces CacheWriteTokens, Cost, and Duration into UsageDetails.AdditionalCounts; it does not surface Model from the usage event at all today. This is a pre-existing .NET gap relative to Python, not something introduced or worsened by this Go PR.

Result: aligned. The Go change brings copilotprovider into parity with the Python github_copilot client's additional_properties["model"] behavior on usage events. The absence of the same field in .NET is a separate, pre-existing upstream inconsistency between Python and .NET — flagging it against this Go PR would be out of scope, since Go should track the strongest available upstream evidence (here, Python) rather than being held to the weaker/incomplete .NET surface. No divergent defaults, no new opt-in surface, no exported API change (AdditionalProperties is an existing field on ResponseUpdate), and the accompanying test (TestConvertToAgentResponseUpdate_UsageEvent_SurfacesModel) directly demonstrates the corrected behavior. Dependency bumps in go.mod/go.sum are incidental and out of scope for parity review.

Generated by Go API Consistency Review Agent · copilot · auto · 32.4 AIC · ⌖ 8.09 AIC · ⊞ 9.2K · ◷