Summary

Separate verbose tool output from the main chat stream, allowing users to monitor agent activity without cluttering the conversation.

Problem

Currently, /verbose on|full pushes tool call information into the same message stream as the conversation. This creates a tradeoff:

• Verbose off: Clean chat, but no visibility into what the agent is doing
• Verbose on/full: Full visibility, but noisy conversation

For users in a "cautious trust" phase — who want to see what their agent does (especially for sensitive operations like SSH, file writes, API calls) —
there is no clean separation.

Proposed Solutions

1. Side panel in Control UI

A dedicated panel (collapsible, resizable) showing tool calls in real-time:

• Tool name + arguments (the actual command, not just "exec")
• Start/end timestamps
• Exit codes / success/failure
• Truncated output (expandable)

2. Per-connection verbose setting

Allow two browser tabs to have independent verbose settings. Currently verbose is session-level, so two webchat tabs share the same setting. If they were
connection-scoped, you could:

• Tab 1: /verbose off (clean chat)
• Tab 2: /verbose full (activity monitor)

3. WebSocket/SSE endpoint for tool activity

A dedicated stream that external tools could subscribe to. Users could tail this in a terminal while chatting in the UI.

4. TUI split pane

For terminal users, a tmux-style split showing chat on one side and tool activity on the other.

Use Case

"If you're SSH'd into a server, I should really see what's going on. Maybe a new terminal window separate to this would be good — one window for the
chat, and another for the verbose info."

This is about trust calibration. Users who give agents access to sensitive systems want visibility without noise.

Workarounds Today

• Use different channels (e.g., Telegram for verbose, webchat for clean) — separate sessions
• Tail /tmp/openclaw/openclaw-YYYY-MM-DD.log — lacks command details, just shows tool=exec

Neither is ideal.