Show HN: Batty – Run a team of AI coding agents in tmux with test gating
TL;DR
Batty is an open-source supervisor layer that coordinates multiple AI coding agents (Claude Code, Codex, Aider) in parallel tmux panes.
Key Points
- Roles are defined in YAML: an Architect plans, a Manager dispatches work, Engineers execute — each agent isolated in its own git worktree.
- Test gating ensures tasks are only marked complete when the code actually compiles and tests pass.
- Batty embeds no model and is not an agent framework — it orchestrates existing agent CLIs using tmux as the runtime.
- The kanban board is a plain Markdown file generated by the bundled kanban-md tool.
Nauti's Take
Batty solves a real problem that anyone who has seriously experimented with multi-agent setups will recognize: parallel work without isolation is chaos. The decision to compose existing tools — tmux, git worktrees, existing agent CLIs — into a supervisor layer rather than building yet another all-in-one framework is both smart and intellectually honest.
The obvious fragility is that the entire coordination logic depends on the stability of each agent CLI's interface — a breaking change in Claude Code or Codex could silently break Batty. Still, for teams wanting productive multi-agent workflows today, this is the most pragmatic approach currently available.