Use the agent name field instead of display_name for collaborator tool name generation in multi-agent configurations

Background / Current Behavior

In watsonx Orchestrate, a multi-agent configuration consists of one router (parent) agent associated with multiple specialist (child) agents as collaborators. We interpret that the router internally registers each collaborator as a "tool" with a name in the format chat_with_collaborator_xxx. The LLM uses this tool name and its associated description to determine which child agent to route to.

The xxx suffix is generated by extracting and concatenating alphabetic sequences from the child agent's display_name.

Problem

When display_name contains only non-Latin characters (Japanese, Chinese, Korean, etc.), no alphabetic sequences can be extracted, and all agents fall back to the identical name chat_with_collaborator_tool.

When multiple collaborators share the same tool name, the last registered one overwrites the others. From the LLM's perspective, only one tool exists — regardless of the question, only the last registered child agent is ever called. No error or warning is raised, making diagnosis extremely difficult.

Proposed Fix (in priority order)

1. Root fix: Use the name field instead of display_name for tool name generation. The name field is guaranteed unique by the platform and language-independent. Using name would eliminate the need for suffix generation logic entirely, resulting in simpler and clearer tool names.

2. Alternative: If changing the naming logic is not feasible, add a note to the multi-agent documentation (particularly alongside the description authoring guidelines) advising that the display_name of each collaborator should contain a unique, identifiable alphabetic string.

Impact

This issue affects all users in Japanese, Chinese, Korean, and other non-Latin character environments. It goes unnoticed in English-speaking environments because agent names naturally contain unique alphabetic characters.