1 comments

  • meller_a 6 hours ago

    I was running Gemini CLI with the conductor format for task tracking, and Claude Code with superpowers skills for implementation — also inspired by OpenClaw's approach. All great tools, but completely siloed. No shared context between agents, no visibility into what each was doing, and if one LLM exhausted mid-task there was no clean handoff to the other.

    I wanted one thing: a unified view where both agents share the same context and I can see what's happening without reading terminal output.

    LaneConductor is that layer — a live Kanban board (localhost:8090) that updates in real-time as your agents work, backed by shared Markdown files that both Claude and Gemini read and write.

    It's built on context-driven development: all state lives in Markdown files (plan.md, spec.md, index.md) rather than in any LLM's context window. If Claude exhausts mid-task, Gemini opens the same files and continues from exactly the same point — no lost work, no restarting from scratch. The files are the shared context and the crash recovery mechanism.

    Four ways to use it: - /laneconductor skill commands inside Claude Code (plan, implement, review, brainstorm) - Instruct Gemini CLI the same way — native conductor format support - lc CLI for track management - Worker mode: background daemon that picks up queued tracks autonomously

    Demo: https://raw.githubusercontent.com/meller/laneconductor/main/...

    Open source, 100% local — Postgres + Vite on localhost. No cloud, no auth, no data leaves your machine.

    https://github.com/meller/laneconductor

    Running this daily across 3 repos with Claude and Gemini simultaneously. Happy to answer questions.