Python: Ignore excluded tool results during compaction

[dk3yyyy]

Motivation & Context

ToolResultCompactionStrategy could summarize messages that had already been marked as excluded when only part of a tool group remained in context. This could restore a large excluded tool result inside the generated summary and increase context token usage instead of reducing it.

This change keeps excluded content out of generated summaries while preserving the existing provenance links between each original message and its summary.

Description & Review Guide

• What are the major changes? Summary content and function-call mappings are now built only from included messages in the selected group. The complete group is still used for provenance IDs and backlinks. A regression test covers excluded-content restoration, token growth, and bidirectional provenance.
• What is the impact of these changes? Partially included tool groups compact without reintroducing excluded payloads. Existing function-call and MCP result handling remains unchanged, and there is no public API change.
• What do you want reviewers to focus on? Please review the boundary between included summary content and full-group provenance, particularly the use of filtered messages for content while retaining all original message IDs for traceability.

Related Issue

Fixes #7387

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.

[dk3yyyy]

@microsoft-github-policy-service agree

[Copilot]

Pull request overview

This PR fixes a compaction bug in the Python core where ToolResultCompactionStrategy could re-materialize tool result payloads that were already marked excluded, causing context token usage to grow during compaction. The update ensures generated tool-result summaries are built only from still-included messages while preserving full-group provenance links for traceability.

Changes:

• Build tool-result summary text and call-id→name mappings from included messages only within a selected tool-call group.
• Preserve provenance/backlinks using the full group’s original message IDs (including excluded messages).
• Add a regression test covering excluded-payload restoration, token growth prevention, and bidirectional provenance.

Reviewed changes

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

File	Description
python/packages/core/agent_framework/_compaction.py	Filters group messages used to build tool-result digests to avoid summarizing excluded tool results while keeping full-group provenance.
python/packages/core/tests/core/test_compaction.py	Adds a regression test ensuring excluded tool results are not restored into the inserted summary and provenance links remain correct.

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_compaction.py	711	58	91%	117–118, 126, 193–194, 212–213, 230, 264, 278, 285, 300–301, 355, 362, 378, 428, 452, 483, 532, 538, 540, 559, 603–608, 620, 704, 706, 721, 766, 828, 955, 961–965, 968–970, 978, 1053, 1055, 1071, 1078, 1083, 1098, 1106, 1205, 1228, 1240, 1336, 1363, 1447
TOTAL	45596	4479	90%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
9416	34 💤	0 ❌	0 🔥	2m 26s ⏱️