I build and maintain open-source AI agent systems. Most of that work currently ships through nanobot, where I'm one of the maintainers.
I care less about making an agent work once than making it keep working: preserving the right state, recovering from provider failures, keeping tool boundaries explicit, and making long-running behavior understandable to the user.
At nanobot, I work across the agent runtime, WebUI and gateway, chat channels, provider integrations, tests, reviews, and security boundaries.
I also maintain skills, a collection of reusable agent workflows for reviewing and fixing pull requests, verification, experiment planning, idea management, and technical writing.
I occasionally write about agent systems, debugging, machine learning, and security at chengyongru.github.io.