Can I use this with existing observability platforms?

Yes, the logging wrapper outputs structured JSON that can be easily ingested by platforms like Datadog, New Relic, or ELK stacks.

How does this skill track Claude API costs?

It uses a built-in estimateCost function that maps current token rates for models like Opus, Sonnet, and Haiku to your actual message usage metrics.

What kind of alerts can I set up with this skill?

It provides recommended thresholds for error rates, p95 latency, daily spending limits, and specific API status codes like 429 (rate limits) and 529 (overloaded).

Does it support tracking Anthropic prompt caching?

Yes, the skill specifically tracks cache_read_input_tokens to help you monitor and optimize the effectiveness of Anthropic's prompt caching features.

Claude API Observability

Name: Claude API Observability
Author: jeremylongshore

byjeremylongshore

•

2,083

•

분석 및 모니터링

Monitors and instruments Claude API calls to track token usage, latency, costs, and error rates in production environments.

This skill provides a standardized framework for monitoring Anthropic Claude API integrations, offering production-ready logging wrappers to capture granular metrics such as input/output tokens, cache performance, and estimated costs. By implementing these observability patterns, developers can set up critical alerts for error spikes, high latency, and budget overruns, ensuring robust and cost-effective AI application performance within their Claude Code workflow.

주요 기능

012,083 GitHub stars

02Configurable alert thresholds for latency and 429/529 errors

03Cache hit/miss monitoring for optimized performance

04Real-time token and cost estimation for Claude models

05Production-ready TypeScript logging wrappers

06Detailed error tracking with request IDs and status codes

사용 사례

01Debugging failed API requests with detailed error logs

02Monitoring production Claude API usage and costs

03Optimizing latency and token consumption in LLM applications

주요 기능

012,083 GitHub stars

02Configurable alert thresholds for latency and 429/529 errors

03Cache hit/miss monitoring for optimized performance

04Real-time token and cost estimation for Claude models

05Production-ready TypeScript logging wrappers

06Detailed error tracking with request IDs and status codes

사용 사례

01Debugging failed API requests with detailed error logs

02Monitoring production Claude API usage and costs

03Optimizing latency and token consumption in LLM applications