Overview

This PR adds three new widgets that display token throughput (tokens per second) for Claude Code sessions:

• InputSpeed: Shows input token processing rate
• OutputSpeed: Shows output token generation rate
• TotalSpeed: Shows combined token throughput

Motivation

Token speed metrics help users understand Claude's response performance and can be useful for:

• Monitoring API performance during sessions
• Identifying slow responses
• Comparing throughput across different models

Implementation Details

Speed Calculation

Speed is calculated by measuring active processing time - the time between user requests and assistant responses:

activeTime = sum of (assistant response timestamp - user request timestamp) for each pair
speed = totalTokens / activeTime

This approach excludes idle time (user thinking, typing) and gives a more accurate representation of Claude's response throughput.

Note: The measured speed includes tool execution time (file reads, bash commands, etc.), so output speed may appear lower than pure token generation speed.

Configuration Example

{
  "lines": [
    {
      "items": [
        { "type": "InputSpeed", "color": "cyan" },
        { "type": "OutputSpeed", "color": "cyan" },
        { "type": "TotalSpeed", "color": "cyan" }
      ]
    }
  ]
}

Example Output

In: 10.9 t/s | Out: 1.3 t/s | Total: 12.2 t/s