Agent Trace Skill

Description

Analyzes agent execution logs to visualize decision flow, tool calls, and identify bottlenecks or errors.

Usage

/agent-trace [incident-id]

Example: /agent-trace INC-2026-02-10-001

What This Skill Does

1. •Reads logs/agent-actions.jsonl for specified incident (or latest)
2. •Reconstructs agent execution flow:
   code

   [10:30:45] Orchestrator: Received incident (payment-service high error rate)
   [10:30:46] Orchestrator → Tool: callDebugger
   [10:30:47]   Debugger: Starting diagnostics
   [10:30:48]   Debugger → Tool: getServiceHealth
   [10:30:49]   Debugger ← Result: unhealthy (1/3 replicas ready)
   [10:30:50]   Debugger → Tool: getRecentLogs
   [10:30:52]   Debugger ← Result: 47 OOMKilled errors
   [10:30:53]   Debugger → Tool: getMetrics
   [10:30:54]   Debugger ← Result: memory at 98%, error_rate 15%
   [10:30:55]   Debugger: Diagnosis complete (confidence: 0.89)
   [10:30:56] Orchestrator ← Result: Memory leak identified
   [10:30:57] Orchestrator → Tool: generateReport
   [10:30:58] Orchestrator: Report generated
   

3. •Highlights:
   • •Tool failures (red)
   • •Low confidence diagnoses (yellow)
   • •Escalations (orange)
4. •Calculates:
   • •Total execution time
   • •Token usage estimate
   • •Number of tool calls
5. •Suggests optimizations if applicable

Example Output

Agent Trace: INC-2026-02-10-001
Execution Time: 13.2s
Tool Calls: 7
Estimated Tokens: ~8,500

Decision Flow:
  ✅ Orchestrator correctly delegated to Debugger
  ✅ Debugger gathered comprehensive evidence (3 tools)
  ✅ High confidence diagnosis (0.89)
  ✅ Report generated successfully

No issues detected.