Claude Code Skill

Attyx ships with a Claude Code skill that gives Claude native control over your terminal — creating splits, sending commands, reading output, and managing sessions, all through the IPC interface.

What it does

When you invoke /attyx in Claude Code, Claude can:

• Create and close tabs and split panes
• Send keystrokes and text to any pane by ID
• Read visible screen content (and scrollback) from any pane
• Manage sessions (create, switch, rename, kill) and route commands to a specific session with -s
• Track and watch the status of AI agents running in other panes (list agents, watch agents)
• Orchestrate multi-pane workflows (e.g. run a server in one pane, test it from another)

The skill gives Claude the full attyx CLI vocabulary with awareness of pane targeting, stable IDs, and best practices for reading output reliably.

Installation

attyx skill install

After installing, restart Claude Code. The skill will appear in the / menu.

Usage

Invoke the skill with /attyx followed by a natural language instruction:

/attyx open a split with htop
/attyx send "hello" to the other pane
/attyx close the other pane
/attyx what's on screen in the right pane
/attyx create a background session for ~/Projects/api
/attyx list sessions

Or just /attyx with no arguments — Claude will ask what you’d like to do.

Examples

Run a dev server and watch its output:

/attyx open a vertical split running "npm run dev", then read its output once it starts

Multi-pane workflow:

/attyx set up a split with python3 on the right, send it "import json; print(json.dumps({'ok': True}))", and read the result

Session management:

/attyx create a new session for ~/Projects/backend called "api"

How it works

The skill file (SKILL.md) is loaded into Claude’s context when you invoke /attyx. It contains:

• The full list of attyx IPC commands (dynamically pulled from attyx --help)
• Rules for pane targeting with stable IDs
• Session management documentation
• Best practices (don’t close your own pane, use \r for Enter, poll for output instead of guessing sleep times)
• Argument interpretation guidelines for natural language

Claude then translates your request into one or more attyx CLI commands, executed via the Bash tool.

Skill file structure

skills/claude/attyx/
  SKILL.md     # Skill definition with frontmatter + documentation
  data.zig     # Embeds SKILL.md for Attyx's built-in help system

The SKILL.md frontmatter defines the skill metadata:

---
name: attyx
description: Control the Attyx terminal via IPC — manage splits, send input, read output, track and watch agent status, orchestrate panes.
allowed-tools: Bash
argument-hint: [action] [args...]
---

Tips

• Always use pane targeting. The skill teaches Claude to capture pane IDs on creation (id=$(attyx split v)) and target panes with -p "$id" instead of juggling focus.
• Long-running commands. For builds, installs, or anything slow, Claude will poll with get-text until the output stabilizes rather than blindly sleeping.
• Don’t close yourself. The skill warns Claude not to run attyx split close without -p, which would close the pane Claude is running in.

Automatic status detection

Separate from the skill, Attyx automatically detects when Claude Code is running in a pane and tracks its state — idle, working, or input (waiting on a permission prompt or question) — shown as a colored status dot on the tab. No setup is required.

This works by merging a small set of lifecycle hooks into Claude Code’s settings.json (in ~/.claude, plus any sibling ~/.claude-* config directories, and honoring CLAUDE_CONFIG_DIR). The merge runs once at startup and is non-destructive — your existing settings and hooks are preserved. The injected hooks only emit a status signal and only do so while running inside Attyx, so they’re a no-op when you launch Claude Code elsewhere.

The same mechanism covers Codex (via ~/.codex/hooks.json), opencode (via a plugin in ~/.config/opencode), and pi.dev (via an extension in ~/.pi/agent/extensions), but only patches them if they’re already set up on your machine.

Other panes — and the skill, or any MCP client — can read this status with attyx list agents and attyx watch agents. See Agent Workflows for how to use it.

MCP server

The skill drives Attyx through the Bash CLI. If you’d rather give a client structured tool calls — Claude Desktop, Claude Code’s MCP support, or any other MCP client — Attyx also ships a built-in MCP server over stdio and HTTP. See MCP Server for transports, setup, and the full tool list.