Worked on the HKUDS/Vibe-Trading repository to enhance the reliability and security of outbound signing in automated trading workflows. Addressed a bug where the Gemini thought_signature was lost during input data conversion in the AgentLoop, which previously caused errors in tool-using turns. The solution involved re-attaching the thought_signature to the converted AIMessage, ensuring outbound requests remained properly signed throughout the process. Implemented regression tests using Python to verify the preservation of signing data and prevent future regressions. Focused on backend development and API integration, the work improved stability and maintained end-to-end signing integrity for secure trading automation.