How should AI agent workspaces handle tool calls and approvals?

I am building Clawke, an open-source native workspace for managing AI agents like OpenClaw, Hermes, and Nanobot.

The problem I am trying to solve: AI agents are becoming more than chatbots. They ask for approvals, call tools, manage tasks, use skills, return files, and expose runtime status. A plain chat or IM-style interface gets awkward quickly.

Clawke takes a native workspace approach: the client renders structured interaction surfaces for tool calls, approvals, tasks, skills, media, and gateway status instead of turning everything into text commands.

It supports iOS, Android, macOS, Windows, and Linux, and the project is open source.

I would love feedback from people building or using agent workflows:

When an AI agent needs confirmation, tool controls, task management, or runtime status, what should the client do better than chat?

GitHub: https://github.com/clawke/clawke

Product Hunt: https://www.producthunt.com/prod...