kelly-agent-observability

Visualize LLM agent telemetry including call volume, latency, and error rates.

4|3|Updated Feb 25, 2026
One-click install
npx skills add https://github.com/mr-kelly/skills --skill kelly-agent-observability
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kelly-agent-observability
Source: https://github.com/mr-kelly/skills/tree/main/skills/kelly-agent-observability
Command: npx skills add https://github.com/mr-kelly/skills --skill kelly-agent-observability

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires hono, @hono/node-server, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill provides a centralized, local dashboard to monitor the performance, cost, and reliability of a fleet of LLM agents, helping you identify bottlenecks and chain-of-thought failures without exposing sensitive data to external monitoring services.

Core Features & Use Cases

  • Fleet Health Monitoring: Track call volume, latency (p50/p95), error rates, and costs across multiple agent archetypes.
  • Trace Debugging: Visualize step-by-step tool call timelines to pinpoint exactly where an agent chain breaks.
  • Human-in-the-loop: Flag specific traces or agents for investigation and maintain a local audit log of interventions.
  • Use Case: Use this dashboard to review the performance of your support triage and booking assistant agents after a deployment to ensure error rates remain within acceptable thresholds.

Quick Start

Start the local observability dashboard by running the launcher script in the app directory and opening the provided local URL in your browser.

Frequently Asked Questions about kelly-agent-observability

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I monitor LLM agent fleet performance locally without sending data to external services?

You can debug complex LLM agent chains through trace-level step analysis, which visualizes step-by-step tool call timelines to pinpoint exactly where an agent chain breaks. This allows you to isolate chain-of-thought failures and identify bottlenecks in specific agent archetypes.

Can I track p50 and p95 latency metrics for multiple LLM agent archetypes?

Yes, you can track p50 and p95 latency metrics, error rates, and costs across multiple agent archetypes using the fleet health monitoring dashboard. It aggregates telemetry data locally to help you review performance and ensure error rates remain within acceptable thresholds after deployments.

Does this observability dashboard support human-in-the-loop intervention logging?

Yes, the observability dashboard supports human-in-the-loop intervention logging by allowing you to flag specific traces or agents for investigation. It maintains a local audit log of these interventions to help you track manual reviews and debugging actions within your agent fleet.

Do I need Hono to run the local LLM agent observability dashboard?

Yes, you need Hono and @hono/node-server installed, as the local observability dashboard operates entirely on a Hono-based server to ensure data privacy and zero external network dependencies. You start the dashboard by running the launcher script in the app directory and opening the provided local URL.

What is the best way to visualize LLM agent telemetry data for debugging chain-of-thought failures?

The best way to visualize LLM agent telemetry data for debugging chain-of-thought failures is using a local dashboard that provides trace-level step analysis and tool call timeline visualization. This approach pinpoints exact failure points in agent chains while keeping sensitive monitoring data internal.