Metrics

Metrics

Display detailed session metrics and agent performance dashboard.

Note: Metrics are ESTIMATES based on available session data. Parallelism and agent usage metrics require conversation-level tracking that may not be fully available to a subagent. Values marked N/A indicate unavailable data.

Usage

• /metrics - Show full metrics dashboard

Algorithm

1. Gather metrics data (in parallel where possible):

   Task Metrics:

   • Read current todo list state (if available)
   • Count by status: total, completed, failed, in-progress, pending

   Parallelism Metrics:

   • Estimate from session history:
     • Max parallel tasks observed in a single batch
     • Average tasks per batch
     • Parallelism efficiency: (parallel batches / total batches) * 100

   Agent Usage:

   • Scan session for agent invocations
   • Count tasks per agent
   • Rank top 5 by task count

   Token Estimate:

   • Approximate based on conversation length
   • Messages * avg tokens per message (rough estimate)

   Error Metrics:

   • Count errors encountered during session
   • Count auto-recovered (retried successfully)
   • Count escalated to user (manual intervention)

   Duration:

   • Session start time (first message) to now
   • Format as hours:minutes

   Learning Metrics:

   • New instincts captured this session (if instincts file updated)
   • Pattern: compare instincts file mtime with session start

2. Display dashboard:

   =============================================
    SESSION METRICS DASHBOARD
   =============================================
   
    Tasks
      Total:           {n}
      Completed:       {n}  ({pct}%)
      Failed:          {n}
      In Progress:     {n}
      Pending:         {n}
   
    Parallelism
      Max parallel:    {n} tasks
      Avg per batch:   {n}
      Efficiency:      {pct}%
   
    Agent Usage (Top 5)
      1. {agent_name}:    {count} tasks
      2. {agent_name}:    {count} tasks
      3. {agent_name}:    {count} tasks
      4. {agent_name}:    {count} tasks
      5. {agent_name}:    {count} tasks
   
    Tokens (estimate)
      Total:           ~{n}K tokens
   
    Errors
      Total:           {n}
      Auto-recovered:  {n}
      Manual:          {n}
   
    Duration
      Session time:    {h}h {m}m
   
    Learning
      New instincts:   {n}
   
   =============================================
   

3. Notes on accuracy:

   • Token estimates are approximate (based on message count heuristic)
   • Parallelism metrics reflect observable behavior, not internal scheduling
   • Agent usage depends on task delegation being traceable in conversation

Notes

• This is a read-only diagnostic; it modifies nothing
• Some metrics may show "N/A" if data is not available
• Best used at end of session for a summary view
• For a quick overview, use /status instead