LynchzDEV

Mission Control — AI skill for Claude Code

AI community

Local web cockpit orchestrating three AI coding engines — Claude Code (tech lead), GLM-5.3-Flash via z.ai (junior fleet), Codex CLI (outside critic).

How to install Mission Control

This entry records only its repository, not the path inside it, so there is no exact command to give. Open LynchzDEV/mission-control and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Mission Control does

Local web cockpit orchestrating three AI coding engines — Claude Code (tech lead), GLM-5.3-Flash via z.ai (junior fleet), Codex CLI (outside critic). Bun + Elysia, browser terminals, live quota, session-flow tracking.

Alternatives in AI

Engineering Notebook — A Bun CLI that ingests Claude Code and Codex session transcripts, generates LLM-powered daily summaries, and s 320 ★
Openteam — A Chrome extension for orchestrating local AI-agent team workflows across web AI services 174 ★
SpexFlow — Node-based workflow for orchestrating codebase context and specifications for AI agents 164 ★

README

Mission Control

**Run multiple AI coding agents like a dev team — from one browser tab.**

Mission Control is a self-hosted cockpit for developers who use more than one AI coding CLI. Point it at your engines — Claude Code, a GLM lane through any Anthropic-compatible endpoint (z.ai coding plan), OpenAI Codex CLI — and get live quota dashboards, one-click job dispatch, real terminals in the browser, and a review queue for everything your agents produce.

![Mission Control — lanes](docs/images/lanes.gif)

Why

If you run several AI coding subscriptions, you know the drill: five terminal tabs, no idea which quota is burning, agent output scattered across worktrees, and "what did the overnight run actually finish?" answered by archaeology. Mission Control puts the whole operation on one screen:

Which engine is healthy, and how much quota is left — before you dispatch, not after the rate-limit hits.
What every task is doing right now — a live session-flow graph tracks each piece of work across engines: spec → implement → cross-review → verify → merged.
Where the results are — completed diffs land in a review queue instead of vanishing into scrollback.

Features

🎛 **Lanes dashboard** — one card per engine: live quota (Claude 5-hour block via ccusage, GLM 5-hour window via the z.ai monitor API, Codex auth state), peak-hour countdown for discounted windows, running jobs and terminals, plus any engine sessions started *outside* the app, detected machine-wide.

🚀 **Headless dispatch** — fire `claude -p` / `codex exec` jobs into any git repo, stream logs live over SSE, kill runaway jobs, get a diff-stat the moment a job lands.

🖥 **Browser terminals** — full interactive TUIs (xterm.js ↔ pty over WebSocket): run Claude Code itself inside the cockpit, detach and reattach with scrollback replay, per-engine environment injected automatically.

🔍 **Review queue** — every completed job with a non-empty diff waits for a human, newest first, with a cop