Skip to main content

Documentation Index

Fetch the complete documentation index at: https://claw-lens.com/llms.txt

Use this file to discover all available pages before exploring further.

The Live Monitor is the real-time view in Claw Lens. It connects to your OpenClaw Gateway over WebSocket and shows every session that has had file activity in the last 3 hours. For each session you can see the agent’s current task, every tool call it has made, and whether it is making progress or has stopped responding.

​
Connection status

The page header shows a colored indicator next to the connection state:
  • Green — Connected: the WebSocket connection to the OpenClaw Gateway is open and the dashboard is receiving live updates
  • Red — Not connected: the connection is down; the page will retry automatically with exponential backoff (starting at 1 second, capping at 30 seconds)
The Live Monitor maintains two update mechanisms simultaneously: a WebSocket subscription for push-based updates and a 5-second polling fallback. This means you continue to see fresh data even if the WebSocket connection drops temporarily.

​
Session status

Livemonitor 1
Each session is classified into one of three states:
StatusColorMeaning
runningGreenThe session file was modified within the last 30 minutes and there are no recent tool failures or error stop reasons
stuckAmberThe session file was recently modified but the agent has had recent tool failures, or the last message ended with stop_reason: error or stop_reason: max_tokens
idleGrayNo file activity for 30 minutes or more; the session is shown at reduced opacity
A glowing pulsing dot on running sessions makes active work visually obvious even when multiple cards are on screen.
A “stuck” session does not necessarily mean the agent has crashed — it means the agent has encountered errors in recent tool calls or hit a stop condition. Check the trace to understand what happened and whether the agent needs intervention.

​
Session cards

Each active session is rendered as a card. The card header shows:
  • Status dot — pulsing green (running), solid amber (stuck), or dim gray (idle)
  • Agent name — the name of the OpenClaw agent
  • Session ID — the full UUID, with the first 8 characters shown in the header for quick identification
  • Status badge — RUNNING, STUCK, or IDLE in matching color
  • Idle time — how long ago the session last had activity (e.g. 2m 14s ago, 1h 5m ago)
  • Last tool — the most recently called tool, if any
A secondary row beneath the header shows the full session ID and two navigation links:
  • View in Sessions — opens the Sessions browser with this session pre-selected
  • View in Timeline — opens the Session Timeline for a detailed message-by-message breakdown

​
Step-by-step trace

Livemonitor 2
Below the header, each card shows a live trace of what the agent is currently doing. The trace is rebuilt from the raw JSONL session file every second. The trace resets to the most recent user message, so you always see the current task rather than the full session history.

​
Trace elements

Shown with a blue YOU label. This is the most recent human message that triggered the current agent turn. Cron prefixes ([cron:...]), Slack routing headers, and system metadata are automatically stripped so you see only the actual task text.
When the model produces extended thinking, a collapsible block appears with a purple thinking label and a character count. Click to expand and read the model’s reasoning. Collapsed by default to keep the trace compact.
Shown with a purple AI label. The assistant’s text response for that turn, italicized.
Each tool invocation appears as a row with:
  • A green ✓ (success) or red ✗ (failure) indicator
  • The tool name in bold
  • A preview of the key argument — Claw Lens checks for command, path, file_path, url, query, expression, pattern, old_string, or content and shows the first non-empty value, truncated to 90 characters
  • The duration of the tool call (e.g. 1.4s, 230ms)
  • A result preview — the first 500 characters of the tool output, shown in a dark code block
When a turn ends with stop_reason: error, the error text is highlighted in a red bordered block. This is the most common indicator that an agent is stuck.
  • Waiting: shown when the session is still open and the last step was a user message or tool call — the agent is waiting for the model to respond
  • Done: shown when the session completed normally with stop_reason: end_turn
The trace shows the last 8 turns of the current task. Older turns are not displayed in the Live Monitor — use the Session Details for detailed messages and session overview, or Session Timeline to inspect tool call timelines and individual tool calls.

​
When no sessions are active

If no sessions have had file activity in the last 3 hours, the page shows:
All quiet Updates automatically when agents become active.
The page continues polling every 5 seconds in the background. As soon as an agent starts a new session, the card appears without requiring a page reload.