[Aditya Kumar (Aditya-XR)]

Motivation & Context

When a Magentic workflow stalls and replans, participant agents are supposed to start fresh. Instead they keep their whole earlier conversation. MagenticAgentExecutor.handle_magentic_reset assigns the new session to self._agent_thread, but AgentExecutor has run the agent with self._session since #3850, so the new session is never used and the old history is sent again after every reset.

Description & Review Guide

• What are the major changes?

  In handle_magentic_reset:

           self._pending_agent_requests.clear()
           self._pending_responses_to_agent.clear()
  +        self._pending_request_order.clear()
           # Reset sessions
  -        self._agent_thread = self._agent.create_session()
  +        self._session = self._agent.create_session()

  _pending_request_order is cleared along with the other pending-request state the handler already resets, so it can't keep IDs of requests that no longer exist.

  The new test test_magentic_reset_gives_participants_a_fresh_session drives a real Agent through a scripted stall, reset and replan, and checks what the agent sends to the model after the reset. Without the change it fails:

  E   AssertionError: agentA kept its pre-reset session history
  E   assert ['before reset', 'reply-1', 'after reset'] == ['after reset']
  

• What is the impact of these changes?

  After a reset, each participant gets a new session, as the class and handler docstrings describe. For the repro in the linked issue, the second call is now ['after reset'] instead of ['before reset', 'reply-1', 'after reset']. Nothing changes for workflows that never reset.

• What do you want reviewers to focus on?

  Whether clearing _pending_request_order belongs here. I added it because the handler already clears _pending_agent_requests and _pending_responses_to_agent, and the order list only tracks IDs from those. I can drop it to keep the change to the session line only.

Related Issue

Fixes #9129

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]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Eduard van Valkenburg (eavanvalkenburg)]

/review

[github-actions]

MAF Automated Review — Iteration 1

Result: No findings
Scope: full PR (1 commit(s)): 1bed6c084b12
Model: gpt-5.6-sol

Overview

The change restores the documented fresh-session invariant by resetting the _session field actually used by both streaming and non-streaming agent runs. It also clears all three correlated pending-request structures, and the end-to-end regression test demonstrates that pre-reset history no longer reaches a participant after replanning. No concrete Critical, High, or Medium regression remains after reconciling the reset lifecycle, checkpoint behavior, provider contracts, and existing safeguards.

Reviewed the supplied pull-request change set across correctness, security/reliability, architecture, and failure behavior.
No publishable findings remained after source verification for this scope.