MC1261573 High
Microsoft Copilot Studio – Tool & Topic Invocation Grader
Summary AI-generated
Microsoft Copilot Studio introduces the Tool & Topic Invocation Grader for verifying tool/topic invocations during evaluations; generally available March 31, 2026.
Written by Azure OpenAI (gpt-4.1) from the text of the post below. It can be incomplete or wrong; the original post is authoritative.
Similar posts
Search for more like this- MC1165633 Microsoft Copilot Studio – Draft with Copilot
- MC1472547 Microsoft Copilot Studio - Custom Grader Library
- MC1471573 Copilot Studio - Use MCP-compliant tools in agent workflows
- MC1391947 Microsoft Copilot Studio - Use MCP-compliant tools in agent workflows
- MC1289783 Microsoft Copilot Studio - Automatic Evaluation from the Test Pane
- MC1325011 Microsoft Copilot Studio - Evaluate the entirety of multi-turn conversations
Original post from Microsoft
We are announcing the Tool & Topic Invocation Grader in Microsoft Copilot Studio. This feature will reach general availability on March 31, 2026.
How does this affect me?
The Tool & Topic Invocation Grader verifies that an agent invoked required tools or topics during an evaluation run. While traditional graders focus on response quality, this grader validates execution behavior and orchestration logic.
Makers define a list of mandatory tools or topics per run. During execution, the grader checks whether the specified items were invoked. Key capabilities of this feature include:
This message is for awareness, and no action is required.
How does this affect me?
The Tool & Topic Invocation Grader verifies that an agent invoked required tools or topics during an evaluation run. While traditional graders focus on response quality, this grader validates execution behavior and orchestration logic.
Makers define a list of mandatory tools or topics per run. During execution, the grader checks whether the specified items were invoked. Key capabilities of this feature include:
- Detection of missing or broken tool calls.
- Validation of routing and orchestration flows.
- Early identification of behavior regressions.
- Clear feedback indicating which tools or topics were not invoked.
This message is for awareness, and no action is required.