TermHive — Design skill for Claude Code

Termhive

**The human-driven multi-agent dashboard.**

Termhive is a web-based control center for coding CLI agents (Claude Code, Codex, Gemini, OpenCode). While autonomous agent platforms hand the steering wheel to AI, Termhive keeps you in the driver's seat: see every agent's screen in one window, coordinate them manually, and intervene the moment something goes sideways.

Think of it as **tmux for coding agents** — with a web UI, project wiki, shared content, and MCP-based agent messaging.

https://github.com/user-attachments/assets/8c95c54b-5c1e-471e-9411-6150993b886b

Why "human-driven"?

Autonomous agents are seductive in a demo. In practice they drift, burn tokens, and silently break things. I shipped two autonomous-agent harnesses before this one — both worked until they didn't, and the "didn't" was expensive to clean up.

Termhive is the opposite bet. You run 2–7 agents in parallel, each doing real work, but **you stay in the loop on every one**. No hidden decisions, no runaway loops, no "come back tomorrow and hope it went well."

The problems it actually solves

• Too many terminal windows — can't find which agent is doing what
• No easy way to share context between agents working on the same project
• No persistent project knowledge — agents forget everything between sessions
• No cross-agent coordination — you end up copy-pasting between windows
• Can't manage agents from mobile / remote — you're stuck at your desk
• Autonomous tools hide too much — when they go wrong, you find out too late

Termhive gives you a browser-based dashboard, a per-project wiki, shared content folders, and native agent-to-agent messaging — all while you stay in control of every prompt.

Features

• Multi-vendor — Claude Code, Codex CLI, Gemini CLI, OpenCode in one UI
• Project organization — Group agents by project, each with its own config
• Terminal streaming — Real xterm.js terminals with live PTY via WebSocket
• **Five termi