Skip to content

Copilot Agent Kit October 2026

Latest

Choose a tag to compare

@cattoolsengineering cattoolsengineering released this 03 Nov 10:26
· 6 commits to main since this release
8c86220

Changes

New feature: Model Assessment

Check how a Copilot Studio agent behaves on a different AI model before you keep that model. Model Assessment captures a baseline on the agent's current model, detects when the model changes, and runs the same tests on the new model using either Copilot Studio evaluations or Copilot Agent Kit tests. Each run gets a regression status of Healthy, Warning or Critical, with per-test results showing what improved, what regressed and what still fails, plus an AI summary of the run. Compare up to three runs side by side with test-by-test results and Copilot Credits, get an AI-assisted model recommendation with a confidence level, and export the comparison to PDF. Learn more about Model Assessment

Agent Review Tool

  • Improved review reliability with traceable findings, clearer progress reporting, background processing, duplicate-run prevention, and recovery options. (#878)

  • Expanded Agent Map for standard Copilot Studio agents. Review results can now show topics, tools, knowledge sources, and connected agents in a single view. (#877)

  • Expanded package review capabilities:

    • Added checks for Microsoft 365 declarative agent packages.
    • Added preview support for reviewing custom skills and their supporting files.
    • Improved recommendations with clearer source evidence and traceability. (#881, #882)
  • Improved the accuracy and usability of review results:

    • More accurate findings and new-review scores.
    • Corrected instruction coverage and skill-file validation.
    • Improved finding navigation and result actions.
    • Corrected PDF exports.
    • Improved installer connection repair. (#873, #874, #875, #876, #879, #880)
  • Introduced a separate managed solution named Agent Review Workflow, which packages the review workflow and installer. The solution provides a guided setup experience for configuring the required connections and review workflow. (#888)

  • Support change: Agent Review Tool is no longer supported in GCC, GCC High (GCCH), or DoD.

  • New reviews of standard agents use the same scoring method as GitHub Copilot agents. These scores can be different from the scores of earlier reviews.

  • Reviews of agents that use skills are more accurate.

For admins: Install the updated Agent Review Workflow solution at the same time as this update.

Agent Inventory

  • Added Shared Editor Count and Shared Viewer Count, providing greater visibility into agent access and collaboration.

  • Temporarily disabled the Usage feature in Agent Inventory.

Agent Value Summary

  • Expanded Agent Value Summary to support GitHub Copilot agents.

Test Automation

  • Expanded testing for Adaptive Card invoke actions. Tests can now validate all supported response types, including plain-text responses and Adaptive Card responses.
  • Removed Share evaluations (Data Donations) from Advanced Testing.

Agent Debugger

  • Added links to the flow and workflow runs executed as part of a conversation.

  • Removed the dependency on Agent Inventory data. Agent and environment details are now loaded according to the environments and agents accessible to the signed-in user.

  • Added the ability to copy a conversation ID directly from the Conversation ID filter.

  • Pre-filtered conversations with more than two turns to ensure users get meaningful conversations to analyze.

  • Improved how questions and branches are shown in Conversation & activities.

PowerShield

  • Improved Sync Connectors reliability by correcting the failure that occurred while fetching connector actions. (#864)

SharePoint Synchronization

  • Corrected file uploads so that SharePoint Synchronization passes the file name, rather than the full SharePoint URL, to x-ms-file-name. This prevents file-name truncation errors for documents with long SharePoint paths. (#810)

  • Corrected synchronization behavior to retain unchanged knowledge items, update modified files in place, and remove only items deleted from the source, avoiding repeated recreation that could leave knowledge items stuck in In Progress. (#842)

  • If a DLP policy blocks the Knowledge source with documents in Copilot Studio connector, that configuration is skipped and logged.

GitHub issues fixed (improvements and issues)

First-time setup instructions

Upgrade instructions