JUN · founder of LIDGE
I run LIDGE, a small lab that ships open-source tools for working with coding agents, and I help teams bring those agents into their own codebases.
A local proxy that lets Codex and Claude Code run on the model you choose. Started June 2026; I maintain it with agents and a small group of outside contributors.
over 16,000 GitHub stars · issues closed in about half a day on average (last 100, Sep 2026)
codexclaw, a Codex plugin that runs every change through plan → audit → build → check → done, with evidence required at each step. It is the operating system behind the issue turnaround above.
A short talk, "Codex, your way": building a dev loop inside Codex with only hooks and skills, and why personal automation rarely needs a backend.
At a consumer-app company, I standardized the documents a coding agent reads first across 128 legacy repositories, with a fail-closed checker. The 37 active repositories passed the first spec; the effect on agents is not measured yet, and I say so.
ima2-gen (local image/video studio for agents), cli-jaw (personal AI assistant), agbrowse (browser automation CLI), aside-codemode (one code call instead of fifty tool calls), design-isms (design reference). This site and lidgeai.com were built with them.
Korea University — B.A. in Economics (in progress)
I do not write most of the code by hand. I plan, split the work, send it to agents, and review what comes back. My job is the loop, not the keystrokes.
Nothing is done because an agent says so. Every step needs fresh output — a build, a test, a screenshot — and I write down what is still unproven.
Small releases, often. When someone files an issue, the goal is a reproduction, a fix and a release, usually the same day.
The fastest way to reach me is email. I read everything.
Open to conversations about Deployed / Forward-Deployed Engineering and developer relations.