Adds a deep_reasoning tool that consults stronger reasoning models (e.g., DeepSeek R1) to produce full reasoning traces for complex problems, helping the agent shape and cross-check its answers.
