trace - SKILL.md Agent Skill

name: trace description: Show agent execution timeline and performance summary argument-hint: "[--filter <tool|agent>] [--last N] [--summary]" disable-model-invocation: true

Agent Flow Trace

[TRACE MODE ACTIVATED]

Objective

Display the flow trace showing how hooks, keywords, skills, agents, tools, and modes interacted during this session.

User Command

$ARGUMENTS

Instructions

Call trace_timeline MCP tool first.
- Default: latest session timeline with no arguments
- If --filter <tool|agent> is provided, pass filter
- If --last N is provided, pass last
Call trace_summary MCP tool after timeline.
Present the timeline first, then the summary.
Highlight:
- Bottlenecks (tools with average duration >5s)
- Mode transitions (execution mode changes)
- Failure clusters (repeated failures by tool/step)
- Flow patterns (keyword → skill → agent/tool chains)
Note runtime wiring when relevant:
- MCP server config lives in mcp_template.json
- Trace server runs via node toolboxes/trace/trace.js

Output Format

## Trace Timeline
<chronological events from trace_timeline>

## Trace Summary
<aggregate metrics from trace_summary>

## Highlights
- Bottlenecks: ...
- Mode transitions: ...
- Failure clusters: ...
- Flow patterns: ...