Engineering
awb avatar

awb

Interactive UI components for Claude Code and AI agents. Create confirmations, checklists, inputs, tables, and views to handle non-blocking interactions and monitoring.

Introduction

Agent Workbench (awb) is a powerful command-line and interface extension designed to enhance the capabilities of AI coding agents like Claude Code. It solves the critical bottleneck of blocking synchronous interactions by providing a non-blocking communication channel. Instead of halting agent execution while waiting for human input, agents can trigger rich UI components such as confirmation dialogs, multi-item checklists, form inputs, dynamic tables, and markdown or code previews. These interactions are queued in a centralized web-based playground, allowing users to review and respond to multiple agent requests asynchronously.

  • Non-blocking interaction flow: Agents continue execution after triggering components, using the awb run and awb wait pattern to retrieve user responses later.

  • Diverse UI component library: Supports confirm, checklist, ask, code, table, markdown, plan-viewer, and html for custom embedded web components.

  • Parallel session management: Monitor multiple AI agent sessions from a single dashboard to track progress and pending requests.

  • Tmux integration: Manage long-running dev servers and background processes directly within the workbench for streamlined developer workflows.

  • Extensible design: Enables custom HTML panels and web components for specialized reporting or interactive feedback loops.

  • Always utilize --title for every component execution to ensure clarity in the playground dashboard.

  • Use awb ui to initialize the playground and monitor active agent sessions.

  • Follow every awb run with awb wait <id> to capture the JSON response generated by the user's interaction.

  • For complex tasks or extended processes, leverage tmux integration to keep background operations persistent and accessible.

  • Ensure awb is correctly installed in the system PATH; if missing, request the user to perform the local CLI installation.

  • Ideal for software engineers, automation architects, and AI developers seeking to improve LLM-based coding agent reliability and interactivity during CI/CD, project scaffolding, or data analysis tasks.

Repository Stats

Stars
13
Forks
2
Open Issues
0
Language
TypeScript
Default Branch
main
Sync Status
Idle
Last Synced
May 3, 2026, 05:15 PM
View on GitHub