trace_skill

Overview

This skill displays an agent flow trace timeline and a concise summary of execution during a session. It surfaces how hooks, keywords, skills, agents, and tools interacted, revealing mode transitions, bottlenecks, and common flow patterns. The output is organized: timeline first, then aggregate statistics for quick diagnosis.

How this skill works

The skill calls two trace MCP tools: trace_timeline to fetch chronological events and trace_summary to compute aggregate metrics. trace_timeline can be filtered by event type or limited to the last N events. trace_summary reports hook counts, detected keywords, activated skills, mode transitions, and tool performance.

When to use it

• Diagnosing why a multi-agent session ran slower than expected
• Understanding which agents and tools were engaged for a given objective
• Auditing how keywords and hooks triggered downstream skills
• Spotting mode transitions that caused unexpected behavior
• Summarizing session activity for post-mortem or reporting

Best practices

• Run the timeline first to get chronological context, then review the summary for hotspots
• Use filters (hooks, skills, agents, keywords, tools, modes) to isolate noisy subsystems
• Limit the timeline with the last parameter when analyzing large sessions to focus on recent events
• Pay special attention to slow tools and repeated retries as indicators of bottlenecks
• Correlate mode transitions with timestamps to spot where execution logic diverged

Example use cases

• Showing a developer which skill triggered an unexpected tool call during a CI run
• Tracing a keyword -> skill -> agent chain that produced a wrong output
• Identifying which agent introduced latency in a parallel execution scenario
• Generating a short report of hook fire counts and active skills after a session
• Filtering the last 50 events to debug a recent failure without noise from earlier activity