Stacks/Coding agent

LangGraph plus CLI coding agents in git worktrees (HN)

Chepko932 combines LangGraph with a homemade coordinator that runs Claude Code, Codex CLI, and Gemini CLI in parallel git worktrees. Inter-agent data is SQLite-structured JSON; the coordinator routes it. The design rule is that humans (or the coordinator) own the task graph, and agents only execute leaves.

Use case

A production multi-agent coding setup in which specialized CLI agents run in parallel, each in its own git worktree, under a central coordinator

Category

Coding agent

Team size

not stated

Orchestrator / framework

LangGraph with a custom orchestrator on top. Agents run as parallel workers (Claude Code, Codex CLI, Gemini CLI), each in its own git worktree. A central coordinator reads structured output and routes work. The author's biggest takeaway: don't let agents pick their own subtasks; define the task graph yourself and let agents only handle the leaf nodes. Letting agents talk to each other directly was 'a mess.'

Memory

Agent-to-agent data flows through SQLite-structured JSON output per task, which the coordinator reads.

Evals

not stated

Observability

not stated

Monthly cost

not stated

Models

Claude CodeCodex CLIGemini CLI

Observability tools

not stated

Other tools

Git worktreesSQLite

Source

https://news.ycombinator.com/item?id=47687389

Chepko932 · Apr 7, 2026