Worked on performance and stability improvements for the Tencent/WeKnora repository, focusing on enhancing caching mechanisms for LLM-driven backend flows. Developed deterministic ordering for tool definitions and listings, ensuring byte-stable outputs that improve cache hit rates and reduce variability. Introduced end-to-end telemetry by adding cached token usage reporting and logging, increasing visibility into prompt-cache effectiveness. Clarified documentation and semantics for explicit-cache providers to prevent misinterpretation and ensure accurate integration. Expanded test coverage to validate deterministic ordering, token usage parsing, and JSON stability. Utilized Go for backend development, API design, and comprehensive testing, delivering more reliable and maintainable caching infrastructure.