What problem does it solve? Running several coding agents (Claude Code, Codex, Gemini, and others) in parallel tmux panes makes it hard to know which agent is blocked, working, or finished, and agents cannot easily read or drive each other's terminals. ## Core Features & Use Cases - Per-agent status detection: Classifies each pane as error, blocked, working, idle, or done using per-agent screen manifests ported from Herdr, with notifications on blocked/error/done transitions. - Agent-to-agent coordination API: Scripts like agent-read, agent-send, agent-run, and agent-wait let one agent monitor and control sibling agents, the shell equivalent of Herdr's socket API. - Zero-command auto-tracking: A zsh hook registers any pane running a known agent binary automatically, while humans interact only through tmux keys (prefix a, then d for dashboard, m for menu, b/e/w/i/f to jump by status). - Use Case: Launch three agents in separate git worktrees, have a coordinator agent wait for the test agent to reach done, read the reviewer's output, and jump to any blocked agent when a notification fires. ## Quick Start Ask the AI to invoke the tmux-agent-herdr-lite skill to install the tmux integration and set up an agent cockpit in your current repository.
