Agent Workbench

Your Command Center for Claude Code.

tui.mp4

Agent Inbox

Running multiple Claude sessions? Agent Workbench gives you a single playground to monitor them all. See which agents are running, thinking, waiting, or idle. Respond to interactions when you're ready—agents wait patiently.

awb ui

• One place for all sessions - See every Claude instance at a glance
• Never miss a request - Interactions queue up until you respond
• Stay in the loop - Know what each agent is doing in real-time

The Problem

Claude Code's built-in AskUserQuestion tool blocks execution until you respond. One question at a time. If Claude needs approval mid-task, everything stops.

The Solution

Agent Workbench is a CLI + Claude Code skill for non-blocking interactions. Claude keeps working while you review and respond in your own time via the web playground.

# Claude runs this (non-blocking)
awb run --title "Deploy" confirm --prompt "Deploy to production?"

# Returns immediately with an ID
# Claude continues working, checks result later
awb wait <id>

• Non-blocking - Claude asks without stopping
• Parallel interactions - Multiple questions queue up in the playground
• Rich components - Diffs, tables, checklists, not just text prompts

Install

claude plugins add-marketplace github:termos-dev/agent-workbench
claude plugins install awb

Then run /awb:init in Claude to configure.

Usage

# Launch the playground to monitor and respond
awb ui

# Show help
awb

Components

confirm ask checklist code table markdown html plan-viewer

Additional UI panels can be embedded via html using <awb-*> web components.

License

MIT