Copilot Declarative Agent always shows confirmation button for every action – how to disable?

Question

Hello Team,

I am facing a usability issue while working with a Copilot Declarative Agent created using the Microsoft 365 Agents Toolkit (VS Code extension).

Issue Summary

Every time I ask a question or request an action through the agent, a confirmation button appears, requiring manual approval before execution. This happens consistently for all interactions.

Steps to Reproduce

1. Create a declarative agent using Microsoft 365 Agents Toolkit in VS Code.
2. Test the agent with Copilot (e.g., in Excel or M365 chat).
3. Ask the agent to perform any action or respond to a query.
4. Observe that a confirmation prompt/button appears before the action is executed.

Expected Behavior

• The agent should execute trusted or low-risk actions directly without requiring confirmation every time.
• Confirmation prompts should be configurable or appear only for sensitive/high-impact actions.

Actual Behavior

• A confirmation button is displayed for every request/action, regardless of its type.
• User must manually approve each step, leading to a fragmented experience.

Impact

• Slows down interaction with the agent.
• Reduces conversational flow and user experience.
• Makes the agent less suitable for rapid or multi-step workflows.

Environment Details

• Framework: Microsoft 365 Agents Toolkit (VS Code extension)
• Feature: Declarative Agent
• Host Application: e.g., Excel / Copilot Chat
• Platform: Web

Questions

• Is this confirmation behavior expected by design?
• Is there a way to disable or reduce confirmation prompts?
• Can confirmation behavior be configured (e.g., only for specific actions or based on risk level)?

Additional Notes

• This behavior appears across all interactions, not tied to a specific action type.
• Looking for possible configuration, manifest setting, or policy-based control to optimize the experience.

---

Thank you for your support.

Answer 1

Hello Patil, Tanisha , 

Yes, this behavior is generally expected by design for Microsoft 365 Declarative Agents. Declarative agents operate within the Microsoft 365 Copilot trust and security model, where users are asked to confirm actions before they execute. This helps prevent unintended actions and ensures the user remains in control of agent operations.

At present, there is no documented manifest setting or configuration option that allows makers to globally disable these confirmation prompts for declarative agents. The confirmation experience is typically enforced by the Copilot host and is not something that can be overridden from the agent definition itself.

If your scenario requires more autonomous execution or finer control over approval flows, you may want to evaluate alternatives such as:

• Copilot Studio agents with agentic capabilities and workflows.
• Custom engine agents or pro-code implementations where workflow behavior can be designed more explicitly.
• Architectures that separate informational responses from action-taking operations.

So, to answer your questions:

Is this expected behavior?
Yes, confirmation prompts for actions in declarative agents are generally expected as part of the Microsoft 365 Copilot safety and user-consent model.

Can confirmation prompts be disabled or reduced?
There is currently no documented setting to disable them globally for Declarative Agents.

1. Can confirmations be configured based on action type or risk level?
   No public configuration is currently documented that allows makers to selectively control confirmations by action type or risk level.

Answer 2

The confirmation behavior is expected for actions and is required by the current validation and Responsible AI guidelines for declarative agents.

From the validation rules for agents:

• For action scenarios, agents must:
  • Seek user confirmation before executing consequential actions.
  • Clearly disclose what data or operation is involved.
  • Provide a confirmation of completion in the form of a card.
  • Require explicit user permission for consequential actions that mutate a system, with:
    • isConsequential = true for plugin actions.
    • readOnlyHint = false for MCP server actions.
• Highly consequential tasks such as bulk delete are discouraged and must not be supported.

This means:

• Confirmation prompts for actions are by design and are part of the required user-disclosure and safety model.
• Declarative agents must pass Responsible AI and validation checks, so removing confirmation for actions that change or expose data is not supported.

Regarding configuration:

• The documentation describes how to customize confirmation text via the Confirmation object in the function’s capabilities (for example, using the body property to make the prompt clearer and more specific), but not how to disable confirmation for actions that are considered consequential.
• For non-mutating, read-only operations, behavior depends on how the underlying action is declared (for example, whether it is marked consequential or read-only). However, the provided guidance focuses on ensuring confirmation is present and clear, not on turning it off.

Based on the available guidance:

• The presence of confirmation prompts for actions is expected and required for validation.
• There is no supported way in the declarative agent manifest or toolkit configuration, as documented here, to globally disable confirmation prompts for actions.
• What can be configured is how the confirmation is presented (for example, customizing the confirmation text), not whether it appears for consequential actions.

---

References:

• Validation guidelines for agents
• Declarative agent schema 1.7 for Microsoft 365 Copilot
• Declarative agents FAQ
• Declarative agents for Microsoft 365 Copilot
• Agents FAQ for Microsoft 365