Worked on enhancing reliability and observability in the crewAIInc/crewAI repository by addressing context propagation issues during parallel tool calls. Focused on ensuring correct event and task tracking across worker threads by implementing independent context copies for each submission using Python’s contextvars.copy_context().run. This approach prevented shared mutable state and maintained accurate inherited context in concurrent executions. Stabilized distributed tracing by preserving OpenTelemetry span parenting and Celery task ID propagation in multi-threaded environments. Leveraged expertise in asynchronous programming, concurrency, and threading to improve debugging efficiency and reduce diagnosis time for context-related issues, contributing to more robust parallel tool invocation workflows.