I am having issue of agent not being published. I have configured a child agent with required inputs in an agent. Previously i was able to publish this kind of configuration. Also the ss below show i can still publish such agents( this one is in different env)
The root cause is a recent, undocumented validation change in Copilot Studio (late Jan / Feb 2026) that introduced stricter publish-time checks for child agents. The platform now performs a static resolvability validation before allowing publish. If a required input is not guaranteed to be resolved at runtime—and the agent produces no observable output—publishing can fail. Previously, a single AI-filled required input was allowed because runtime resolution was trusted; now the system requires clearer guarantees.
This change was introduced to prevent runtime agent chaining failures and unclear input provenance, especially with autonomous/background agents. The breaking combination in your case is: child agent + required input + AI-filled source + no parent binding + no output.
Recommended fixes include:
- Add a dummy output (cleanest fix) and respond with it to satisfy validation.
- Make the input optional and handle inference in instructions.
- Explicitly bind the input from the parent agent.
- (Less ideal) Add a second required input.
This issue appears only in newer or updated environments due to staged rollout of the new validation logic.
TL;DR: Not a configuration mistake—this is a new publish-time validation rule. The simplest fix is to add an output or change completion behavior so the agent produces observable output
Reference Document0-
1.Add a child agent - Microsoft Copilot Studio | Microsoft Learn
2.Using Inputs and Outputs in Child and Connected Agents | The Custom Engine
3.Microsoft Copilot Studio troubleshooting | Microsoft Learn

Report
All responses (
Answers (