Can it detect if an agent is failing consistently?

Yes, it tracks total runs versus success counts. You can use these metrics to identify agents with success rates below 70% and trigger alerts or workflow changes.

Where are the metrics stored?

All performance data is stored locally in a standardized JSON schema within the .claude/agent-performance.json file for easy inspection and programmatic analysis.

How does this skill help reduce AI costs?

By tracking the cost-per-task across different models, it identifies opportunities where cheaper models (like Grok or smaller Claude versions) can replace expensive ones without sacrificing quality.

What does the Performance Tracking skill do?

It monitors and logs the performance of AI agents, skills, and models to help users optimize success rates, response times, and total costs during development.

Is it compatible with multi-model workflows?

Absolutely. It is specifically designed to work with plugins like multimodel to provide the data needed for intelligent model routing and tier distribution.

Performance Tracking

Name: Performance Tracking
Author: MadAppGang

byMadAppGang

•

분석 및 모니터링

Monitors and optimizes AI agent success rates, model latency, and operational costs through automated metrics collection.

Performance Tracking provides a robust framework for quantifying the effectiveness of your AI workflows by logging detailed metrics for agents, skills, and models. By capturing data on success rates, execution times, and actual spend, it enables data-driven decisions for routing tasks to the most efficient models and identifying underperforming agents. This skill transforms your development environment into an observable system, allowing for continuous optimization of prompt engineering and model selection based on historical evidence rather than intuition.

주요 기능

01Real-time success and failure rate monitoring for specific AI agents

023 GitHub stars

03Granular model performance tracking including latency and cost per task

04Automated routing feedback loops to optimize model tier selection

05Historical execution logs for trend detection and performance benchmarking

06Skill activation analysis to determine tool effectiveness and usage patterns

사용 사례

01Identifying and debugging agents with low completion rates or high failure frequencies

02Benchmarking different LLM providers to find the best cost-to-performance ratio for specific coding tasks

03Optimizing routing logic to use cheaper models for high-confidence repetitive tasks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/magus performance-tracking

For use in Claude.ai and ChatGPT

주요 기능

01Real-time success and failure rate monitoring for specific AI agents

023 GitHub stars

03Granular model performance tracking including latency and cost per task

04Automated routing feedback loops to optimize model tier selection

05Historical execution logs for trend detection and performance benchmarking

06Skill activation analysis to determine tool effectiveness and usage patterns

사용 사례

01Identifying and debugging agents with low completion rates or high failure frequencies

02Benchmarking different LLM providers to find the best cost-to-performance ratio for specific coding tasks

03Optimizing routing logic to use cheaper models for high-confidence repetitive tasks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/magus performance-tracking

For use in Claude.ai and ChatGPT