web
You’re offline. This is a read only version of the page.
close
Skip to main content

Announcements

News and Announcements icon
Community site session details

Community site session details

Session Id :
Power Platform Community / Forums / Copilot Studio / explanation_of_tool_ca...
Copilot Studio
Suggested Answer

explanation_of_tool_call error

(0) ShareShare
ReportReport
Posted on by 6

Hi everyone,

Has anyone managed to successfully handle explanation_of_tool_call and tool_execution_message when using the GPT-5 model?

Please don’t paste generic AI-generated answers — I’ve already tried pretty much everything (additional instructions, setting up a separate thread for the issue, etc.), but the error still persists.

Do you know if there’s any information on when Microsoft plans to fix this? I’ve noticed this issue has been coming up for several months now and still hasn’t been resolved.

Thanks in advance for any real insights.

Categories:
I have the same question (0)
  • Sunil Kumar Pashikanti Profile Picture
    2,318 Moderator on at
     
    Practical reality right now:
    If you need a fix today: Downgrade the agent to GPT‑4.1 (or another stable model). It’s currently the most stable option and avoids the explanation_of_tool_call issue altogether.
    OR
    If you raelly need GPT‑5’s reasoning: Use a two‑model setup
    – GPT‑5 --> reasoning & planning only (no tools)
    – GPT‑4.1 (or another stable model) --> executes tool calls

    Why this helps:
    When explanation_of_tool_call shows up, the failure is happening in the orchestration layer, after the model has already responded. At that point, no amount of prompt engineering will help, the runtime isn’t sanitizing internal message types correctly.

    Until the platform side is updated, architectural workarounds like this are the only reliable option I’ve seen work.
     
     
  • Suggested answer
    Ashlesha-MSFT Profile Picture
    Microsoft Employee on at
    This is a known platform issue confirmed by multiple Microsoft staff members. The Copilot Studio orchestration runtime generates internal reasoning about tool and sub-agent calls, and a bug causes this metadata to leak into the user-facing chat as explanation_of_tool_call JSON. This is not a configuration error on your end.
     
    Temporary workarounds while waiting for the platform fix:
    1. Add to your agent's instructions: "Do not output internal reasoning, explanations of tool calls, or diagnostic JSON fields. Only return the final answer intended for the end-user."
    2. If feasible, switch from GPT-5 to GPT-4 — the issue occurs more frequently on GPT-5
    3. For custom topics: map tool output to a clean text variable (dialog.topicResponse = toolResult.responseText) instead of passing the raw tool output
    Recommended next step: Open a support ticket through the Power Platform Admin Center → Help + Support. Describe the explanation_of_tool_call leak and ask whether your environment has been moved to the updated orchestration runtime. This is the fastest way to get visibility on when the fix reaches your tenant.

Under review

Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.

Helpful resources

Quick Links

Season of Sharing Community Challenge Launch!

Jump in, show your community spirit, and win prizes!

Kudos to our 2025 Community Spotlight Honorees

Expanding mentorship, skilling, and AI innovation

Congratulations to the May Top 10 Community Leaders!

These are the community rock stars!

Leaderboard > Copilot Studio

#1
Valantis Profile Picture

Valantis 277

#2
11manish Profile Picture

11manish 206

#3
sannavajjala87 Profile Picture

sannavajjala87 156 Super User 2026 Season 1

Last 30 days Overall leaderboard