Herdctl — Design skill for Claude Code

herdctl — NTM Control Plane (tmux / Herdr)

Here's a brief overview, starting with a real example:

export NTM_BACKEND=herdr        # switch to herdr (default = tmux)
herdctl spawn api --cc=2 --cod=1
herdctl list
herdctl status api
herdr                            # native herdr TUI
herdctl send api --all "refactor the auth layer."

herdctl is a **dual-backend** control plane for multi-agent AI coding sessions. It orchestrates Claude Code, Codex, Antigravity, Gemini, Cursor, and other coding agents on either `tmux` (the original NTM backend) or [`herdr`](https://herdr.dev) (herdr backend with richer TUI, sidebar agent status, and persistent panes).

The NTM orchestration semantics — spawn, send, work triage, Agent Mail, safety policy, checkpoints, pipelines — are **shared across both backends**. Only the substrate (session/pane/layout/attach) changes. See [`FEATURES.md`](FEATURES.md) for current herdr backend parity.

TL;DR

The Problem

Running several coding agents in parallel is easy to start and annoying to sustain. A plain terminal multiplexer gives you panes, but it does not give you durable coordination, work selection, safety policy, approvals, history, replayable automation surfaces, or a shared control model that both humans and agents can use.

The Solution

herdctl gives you a single local system for:

• spawning labeled multi-agent sessions
• sending work, interrupts, and follow-ups across panes
• triaging what to do next with br and bv
• coordinating agents with