Kun Chen builds open-source tools for coordinating and reviewing the work of coding agents. His projects include firstmate and no-mistakes. He previously led Rovo Dev engineering at Atlassian and held engineering roles at Microsoft and Meta.[1][3][4][5]
Start with firstmate →
Career
Atlassian’s conference biography identifies Chen as the lead principal engineer directing Rovo Dev, an AI agent for software-development tasks. It also records earlier engineering roles at Microsoft and Meta. His current GitHub biography describes all three companies as former employers. His public work now includes open-source agent tools and practical explanations of the development environment he uses.[1][2][3]
Contributions
Coordinating a crew of coding agents
Firstmate gives a developer one agent to talk to while other agents work on separate tasks. Its instructions and helper scripts organize isolated Git worktrees, supervise progress and return completed changes or investigation reports. It runs through existing terminal coding tools rather than supplying a new model. This makes it relevant when the difficulty is tracking several agent sessions, their context and unfinished work. Using it requires familiarity with Git and a supported coding-agent environment.[4]
Reviewing changes before publication
No-mistakes places a local Git proxy between a developer’s work and the real remote repository. A push starts an AI-assisted review and validation process in a separate worktree; the initial change is forwarded only after the configured local checks pass. The tool can apply fixes or present findings for a person to resolve. Its name is not a guarantee of correctness: the result still depends on the checks, models and review decisions used.[5]
Ideas
Chen’s kun repository packages his published engineering opinions and tool notes as an agent skill. Invoking it gives an assistant access to a maintained personal knowledge base. It is a way to reuse one engineer’s working approach, with the limits that implies: these are Chen’s views and practices, not a consensus standard. His Linktree also gathers demonstrations, writing and project links for following how that approach develops.[6][2]
Tools, writing & talks
Tools
Start here
firstmate
github.comCoordinates multiple coding agents through one conversation, with isolated worktrees and supervision. Read the workflow overview before trying it with an existing coding-agent setup.
Read the README to understand how Chen coordinates several coding agents through one lead agent. Try it on a small repository if you are comfortable with Git worktrees and terminal coding tools.
Repository owner and maintainer[4]- ProgramBench
github.comBenchmark testing whether AI agents can rebuild working programs from only compiled binaries and documentation.
Catalog workOriginal source - no-mistakes
github.comRuns review and validation in an isolated worktree before forwarding changes to the configured Git remote.
Repository owner and maintainer[5]