During April 2026, Yuhan Lei contributed to the NousResearch/hermes-agent repository by addressing a bug affecting token usage accuracy in the ACP Agent. Lei enhanced backend reliability by implementing conditional logic in Python to detect and map token-related keys from top-level result fields into the Usage object during conversation runs. This approach ensured that token usage data was accurately extracted and populated, directly supporting improved analytics and cost tracking. Lei validated the solution with dedicated testing, demonstrating proficiency in API integration and backend development. The work focused on data integrity and traceability, resulting in more reliable usage metrics for reporting and billing.