The primary reason for this failure is the disable-model-invocation: true setting in the Summarizer's YAML frontmatter.In the GitHub Copilot Agent configuration framework, when an orchestrator agent automates a multi-agent workflow, it relies on the base LLM model to agentically trigger and delegate tasks to its subagents.
Blocks Subagent Delegation: Setting disable-model-invocation: true instructs GitHub Copilot to completely prevent the model from automatically invoking or calling this agent as a subagent.
Requires Manual Intervention: When this property is true, the agent can only be triggered via a direct manual request by the user (such as explicitly picking it from a chat menu or a slash command). Because user-invocable is also set to false, it becomes completely unreachable in this workflow.
Contradicts Orchestration: Even though Orchestrator explicitly registers Summarizer in its agents property, the underlying model respects the disable-model-invocation: true safety/routing block and refuses to spin up the subagent loop for it.
Reference:
https://docs.github.com/en/copilot/how-tos/copilot-sdk/features/custom-agents