[dotnet-port-fixes] Aggregate loop usage across iterations

[Michelle Clayton (michelle-clayton-work)]

Summary

Align the Go loop harness with the LoopAgent portion of microsoft/agent-framework#7539 so NonStreamingReturnsLastResponseOnly still reports usage from every loop iteration. The loop middleware now accumulates usage across all iterations, removes per-iteration usage from the buffered final response, and reattaches the aggregated usage to the surfaced last response. I also added a regression test that exercises summed token counts and merged AdditionalCounts.

Ported .NET PRs

• .NET: Aggregate usage across looping agents and chat clients agent-framework#7539 - aggregate usage across looping agents and chat clients

Upstream commit referenced: ec32e86646469585bf6d7be83b343c88bf606950 from microsoft/agent-framework. This PR ports only the LoopAgent behavior that maps to agent/harness/loop.

Breaking Changes

No. This corrects internal usage accounting for existing loop runs without changing exported Go API.

Tests and Examples

• Ran go test ./agent/harness/loop
• Added regression coverage in agent/harness/loop/loop_test.go for aggregated usage in last-response-only non-streaming runs
• No example changes were needed for this internal behavior fix

Notes

This intentionally scopes the upstream port to the Go loop harness slice of microsoft/agent-framework#7539. The broader .NET PR also updated other looping components that do not map to this narrow Go change set.

Generated by .NET to Go Fixes and Test Porting Agent · gpt54 · 250.2 AIC · ⌖ 14.4 AIC · ⊞ 24.2K · ◷

Closes #847

[Copilot]

Pull request overview

This PR updates the loop harness’s non-streaming “return last response only” behavior so token usage is accumulated across all loop iterations, while still surfacing only the final iteration’s response content. It aligns Go behavior with the referenced upstream LoopAgent change and adds a regression test to prevent usage undercounting.

Changes:

• Accumulate UsageDetails across loop iterations when NonStreamingReturnsLastResponseOnly is enabled.
• Strip per-iteration UsageContent from the buffered final iteration updates, then attach the aggregated usage to the surfaced final response.
• Add a regression test asserting summed token counts and merged AdditionalCounts.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
agent/harness/loop/loop.go	Aggregates usage across iterations and reattaches it to the final surfaced response update.
agent/harness/loop/loop_test.go	Adds regression coverage ensuring usage totals are summed across iterations in last-response-only mode.

---

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

[github-actions]

Parity Review — ✅ Approved

This PR ports the loop-agent usage aggregation behavior from upstream microsoft/agent-framework#7539 into the Go loop harness.

Scope: Internal implementation only — all new functions (responseUpdatesWithAggregatedUsage, stripUsageFromResponseUpdate, hasResponseUpdateMetadata, isZeroUsageDetails) are unexported. No exported Go API surface changed.

Parity finding: The fix aligns Go behavior with the .NET LoopAgent change in upstream commit ec32e86, which ensures NonStreamingReturnsLastResponseOnly reports accumulated usage across all loop iterations rather than only the final iteration's usage. The Go implementation faithfully reproduces this semantic: it accumulates UsageDetails across iterations, strips per-iteration usage from buffered updates, and reattaches the aggregate to the surfaced last response.

Label actions: public-api-change label not added (no exported API changed). parity-approved added.

Generated by Go API Consistency Review Agent for #862 · sonnet46 · 16.2 AIC · ⌖ 5.57 AIC · ⊞ 5.7K · ◷

[Quim Muntal (qmuntal)]

I don't like how this PR ported the .NET implementation. Closing, will port it manually.