wtf is l3-simulator?

L3 Simulator is a tool that manages AI coding agents for you on top of GitHub issues. Instead of you manually asking an AI assistant to work on a task, you label a GitHub issue as ready and the tool claims it, dispatches an AI coding agent such as Claude Code or Codex into an isolated copy of your repository, and has it write code and open a pull request. From there it also watches over that pull request, reacting to failing checks, new review comments, or merge conflicts by sending the agent back in to fix things. GitHub labels are the whole system's memory. Labels like ready, in progress, blocked, and in review describe exactly where each issue stands, so if the tool crashes or restarts it can figure out the current state again just by reading labels and pull requests. You can watch and interact with everything either through a terminal dashboard or directly through the normal GitHub issues page. Importantly, the tool never merges anything itself. Merging and final code review stay entirely your responsibility. The agents are blocked from running git push or gh commands that would merge code, so they can open as many pull requests as they like, but shipping the result is always a human decision. The workflow supports optional planning steps, where an agent proposes a plan and waits for your approval before writing any code, and it can even handle situations where one repository's task turns out to depend on work in another repository, automatically filing the follow up issue there. It runs as a command line tool built with Node.js version 22 or newer, and needs the GitHub CLI and at least one of Claude Code or Codex installed to actually do the coding work.