How does it improve task routing?

It provides a feedback loop to routing skills, ensuring tasks are sent to tiers and agents with the highest historical success rates for that specific task type.

Does it track performance across different LLM providers?

Yes, it captures metrics for all models used in your environment, including latency and success counts for providers like Anthropic and xAI.

What happens if an agent's performance drops?

The skill tracks success rates and can trigger alerts or recommendations when an agent's success rate falls below a defined threshold, such as 70%.

Where is the performance data stored?

All metrics are stored locally in a standardized JSON format within the .claude/agent-performance.json file for easy access and privacy.

Can this skill help reduce my API costs?

Yes, it tracks cost-per-task and model efficiency, allowing you to identify where cheaper models can replace expensive ones without sacrificing quality.

Performance Tracking

Name: Performance Tracking
Author: MadAppGang

byMadAppGang

•

230

•

Analytics & Monitoring

Monitors agent success rates, model latency, and execution costs to optimize AI workflow efficiency.

The Performance Tracking skill transforms Claude Code workflows into data-driven systems by capturing granular metrics across agents, skills, and LLM models. It enables developers to monitor success rates, track model latency, and calculate real-time costs, providing the insights needed to optimize task routing and identify underperforming components. By maintaining a historical log of executions and success correlations, it creates a continuous improvement loop that ensures the most efficient and cost-effective models are utilized for every task.

Key Features

01Detailed model performance analytics including latency and cumulative USD cost tracking

02Skill activation monitoring to identify correlations between specific tools and task success

03Automated performance degradation alerts to identify agents needing refinement

04Intelligent routing feedback loop that informs tier selection based on historical data

05230 GitHub stars

06Granular agent success and failure tracking with confidence score averaging

Use Cases

01Improving workflow reliability by routing complex tasks to historically high-performing agents

02Debugging agent behavior by analyzing execution history and failure patterns

03Optimizing AI spending by identifying cheaper model alternatives with high success rates

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/claude-code performance-tracking

For use in Claude.ai and ChatGPT

Download Skill