Build AI chat interfaces using ai-elements components — conversations, messages, tool displays, prompt inputs, and more. Use when the user wants to build a chatbot, AI assistant UI, or any AI-powered chat interface.
日本語の概要は準備中です。原文の説明を表示しています。
Deep expertise in cmux — the terminal multiplexer with native browser views. Use when managing panes, reading terminal output, sending keystrokes, opening browser views, or manually testing web UIs and TUIs inside cmux. Triggers on: cmux, open a browser pane, split terminal, read screen, send keys, test this UI in cmux, preview in cmux.
インストール方法を見るインストールする前に、エージェントに与えられる指示の中身を確認できます。
cmux manages terminal panes and browser views through a Unix socket CLI.
You are already running inside cmux — your current pane has env vars
CMUX_WORKSPACE_ID and CMUX_SURFACE_ID set automatically.
window → workspace(s) → pane(s) → surface(s)
↑
terminal OR browser
cmux uses short refs by default: workspace:1, pane:2, surface:3.
Most commands auto-target the caller's workspace/surface via env vars.
# Discover current layout
cmux identify # What workspace/pane/surface am I in?
cmux list-panes # Panes in current workspace
cmux list-workspaces # All workspaces
# Create a new terminal pane (split from current)
cmux new-pane --direction right
cmux new-pane --direction down
# Create a browser pane
cmux new-pane --type browser --direction right --url http://localhost:3000
# Focus a pane
cmux focus-pane --pane pane:3
# Resize
cmux resize-pane --pane pane:2 -R --amount 20 # grow rightward
cmux resize-pane --pane pane:2 -D --amount 10 # grow downward
# Read what's on screen in another pane
cmux read-screen --surface surface:2
cmux read-screen --surface surface:2 --scrollback --lines 100
# Send text (like typing)
cmux send --surface surface:2 "npm run dev"
cmux send --surface surface:2 $'\n' # press Enter
# Send special keys
cmux send-key --surface surface:2 Enter
cmux send-key --surface surface:2 Up
cmux send-key --surface surface:2 Ctrl+C
# Open URL in an existing browser surface
cmux browser --surface surface:3 navigate http://localhost:3000
# Snapshot the DOM (interactive elements)
cmux browser --surface surface:3 snapshot --interactive
# Click, fill, type using CSS selectors
cmux browser --surface surface:3 click "button.submit"
cmux browser --surface surface:3 fill "input[name=email]" "test@example.com"
cmux browser --surface surface:3 type "input[name=search]" "query"
# Press keys, scroll
cmux browser --surface surface:3 press Enter
cmux browser --surface surface:3 scroll --dy 500
# Read state
cmux browser --surface surface:3 get url
cmux browser --surface surface:3 get title
cmux browser --surface surface:3 get text "h1"
# Wait for conditions
cmux browser --surface surface:3 wait --selector ".loaded"
cmux browser --surface surface:3 wait --load-state complete
cmux browser --surface surface:3 wait --text "Dashboard"
The pattern: start the dev server in a terminal pane, open a browser pane pointing at it, then interact via snapshot → act → re-snapshot.
# 1. Start dev server in a new pane
cmux new-pane --direction down
# note the surface ref from output, e.g. surface:2
cmux send --surface surface:2 "npm run dev"
cmux send-key --surface surface:2 Enter
# 2. Wait for server to be ready
# (read screen until you see the ready message)
cmux read-screen --surface surface:2
# 3. Open browser pane pointing at dev server
cmux new-pane --type browser --direction right --url http://localhost:3000
# note the browser surface ref, e.g. surface:3
# 4. Snapshot → interact → re-snapshot loop
cmux browser --surface surface:3 snapshot --interactive
cmux browser --surface surface:3 click "nav a[href='/settings']"
cmux browser --surface surface:3 wait --load-state complete
cmux browser --surface surface:3 snapshot --interactive
# 5. Fill a form
cmux browser --surface surface:3 fill "input[name=username]" "testuser"
cmux browser --surface surface:3 fill "input[name=password]" "secret"
cmux browser --surface surface:3 click "button[type=submit]"
cmux browser --surface surface:3 wait --text "Welcome"
cmux browser --surface surface:3 snapshot --interactive
--interactive returns only interactive elements — much smaller output--compact reduces whitespace for smaller snapshots--selector "main" scopes to a region of the page--max-depth 5 limits DOM tree depthThe pattern: launch the TUI in a new pane, read-screen to observe, send/send-key to interact.
# 1. Launch the TUI
cmux new-pane --direction right
# note surface ref, e.g. surface:2
cmux send --surface surface:2 "my-cli-wizard"
cmux send-key --surface surface:2 Enter
# 2. Read → interact → read loop
cmux read-screen --surface surface:2
cmux send-key --surface surface:2 ArrowDown
cmux send-key --surface surface:2 ArrowDown
cmux send-key --surface surface:2 Enter
cmux read-screen --surface surface:2
# 3. Type text input
cmux send --surface surface:2 "my-project-name"
cmux send-key --surface surface:2 Enter
cmux read-screen --surface surface:2
# 4. Test cancellation
cmux send-key --surface surface:2 Ctrl+C
cmux read-screen --surface surface:2
# 5. Clean up
cmux send-key --surface surface:2 Ctrl+C
read-screen (not read-screen --scrollback) for current viewport--lines 50 with --scrollback to limit output sizeCtrl+C to kill the TUI when donecmux interactions are context-expensive. Minimize round-trips:
--interactive and --compact for browser snapshots--selector when you only care about part of the pagecmux provides sidebar status, progress, and logging for visibility:
cmux set-status "phase" "testing" --icon "🧪"
cmux set-progress 0.5 --label "Running tests..."
cmux log "Found 3 issues" --level warn
cmux clear-progress
For full command details, see:
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Build AI chat interfaces using ai-elements components — conversations, messages, tool displays, prompt inputs, and more. Use when the user wants to build a chatbot, AI assistant UI, or any AI-powered chat interface.
日本語の概要は準備中です。原文の説明を表示しています。
Search the live web via Perplexity Search API. Use when you need current documentation, release notes, vendor pages, news, domain-constrained web search, or date/recency filtering. Not for local codebase search or stable docs already in context.
日本語の概要は準備中です。原文の説明を表示しています。
Chrome DevTools CLI for browser automation via shell commands. Use when interacting with web pages from the command line — navigating, clicking, filling forms, inspecting console/network, taking screenshots, or extracting page content. Triggers on: browse a page, automate Chrome, inspect console, check network requests, take a screenshot, fill a form, click a button.
日本語の概要は準備中です。原文の説明を表示しています。
Uses the chrome-devtools-axi CLI for browser automation, accessibility-tree snapshots, console and network inspection, screenshots, Lighthouse audits, and performance traces. Use when interacting with Chrome from the shell, especially when the user mentions chrome-devtools-axi, AX snapshots, browser debugging, or DevTools automation from the command line.
日本語の概要は準備中です。原文の説明を表示しています。
Uses the gh-axi CLI for GitHub shell operations: issue, pull request, workflow run, release, repo, search, and API tasks. Prefer this over regular `gh` for GitHub reads and simple mutations when an agent needs compact, structured, suggestion-rich output. Triggers on: gh, GitHub CLI, github issue, github pr, pull request, workflow run, github release, gh api, repo inspection, list PRs, view issue, check workflow runs, inspect repo, GitHub shell operations.
日本語の概要は準備中です。原文の説明を表示しています。
Work with Graphite (gt) for stacked PRs. Use when creating, navigating, submitting, or managing PR stacks, or when the user mentions gt, Graphite, or stacked diffs.
日本語の概要は準備中です。原文の説明を表示しています。