[westey (westey-m)]

Motivation & Context

A harness agent that delegates work to background agents can finish its own turn while those background tasks are still running, stopping before their results are available. This adds a LoopEvaluator that keeps re-invoking the harness agent until every background task has reached a terminal state, so the agent cannot declare the work done while tasks are still in flight. It complements the existing TodoCompletionLoopEvaluator and is opt-in.

Description & Review Guide

• What are the major changes?
  • BackgroundAgentsProvider gains a public GetIncompleteTasks(AgentSession?) that refreshes in-flight task state (finalizing just-finished tasks) and returns only the tasks that are still Running.
  • New BackgroundTaskCompletionLoopEvaluator (+ BackgroundTaskCompletionLoopEvaluatorOptions): resolves the BackgroundAgentsProvider at evaluation time from the looped agent via GetService, applies in every mode (no mode gating), and continues with a customizable feedback message while running tasks remain. The feedback template supports {incomplete_task_count} and {incomplete_tasks} placeholders and matches the wording of the parallel Python implementation.
  • Unit tests covering provider resolution, running→continue (with task list/count), terminal→stop, no-tasks→stop, custom template, and GetIncompleteTasks returning only running tasks.
• What is the impact of these changes?
  • Purely additive and opt-in. Existing behavior is unchanged unless a consumer adds this evaluator to a LoopAgent.
• What do you want reviewers to focus on?
  • Provider resolution via the GetService decorator chain (no constructor wiring).
  • The "only Running is incomplete" semantics — Completed, Failed, and Lost are all terminal (a Lost task can never be waited on, so blocking on it would loop forever).

Related Issue

Fixes #6733

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue (see Related Issue above).
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.

[Copilot]

Pull request overview

Adds a new opt-in .NET harness LoopEvaluator that prevents a looped harness agent from declaring completion while background-agent tasks are still running, by re-invoking until all tasks reach a terminal state.

Changes:

• Adds BackgroundTaskCompletionLoopEvaluator (+ options) that resolves BackgroundAgentsProvider via GetService and continues the loop while any tasks remain Running, with configurable feedback templating.
• Extends BackgroundAgentsProvider with GetIncompleteTasks(AgentSession?) to refresh/finalize in-flight tasks and return only still-running tasks.
• Introduces unit tests covering evaluator behavior and the provider’s “running-only” incomplete semantics.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
dotnet/src/Microsoft.Agents.AI/Harness/Loop/BackgroundTaskCompletionLoopEvaluator.cs	New loop evaluator that continues while background tasks are running and emits templated feedback.
dotnet/src/Microsoft.Agents.AI/Harness/Loop/BackgroundTaskCompletionLoopEvaluatorOptions.cs	Options type for configuring the evaluator’s feedback template.
dotnet/src/Microsoft.Agents.AI/Harness/BackgroundAgents/BackgroundAgentsProvider.cs	Adds GetIncompleteTasks that refreshes task state and returns only Running tasks.
dotnet/tests/Microsoft.Agents.AI.UnitTests/Harness/Loop/BackgroundTaskCompletionLoopEvaluatorTests.cs	New unit tests for evaluator/provider behavior and feedback formatting.

[github-actions]

Automated Code Review

Reviewers: 5 | Confidence: 92% | Result: All clear

Reviewed: Correctness, Security Reliability, Test Coverage, Failure Modes, Design Approach

---

Automated review by westey-m's agents