Does it support prompt caching?

Yes, it is designed to identify and enable prompt caching for system prompts exceeding 1,024 tokens to maximize efficiency and cost savings.

How does it prevent runaway costs?

It implements circuit breakers that halt API calls once specific cost, token, or error thresholds are breached within a set timeframe or per-user budget.

What models does this skill support?

It primarily optimizes for the Anthropic Claude family (Haiku, Sonnet, Opus) using the Anthropic SDK within Supabase Edge Functions.

Do I need external tools for the cost dashboard?

No, it utilizes your existing Supabase Postgres database to log and query telemetry natively, avoiding the need for third-party monitoring subscriptions.

What is shadow testing in this context?

It routes a percentage of traffic to a challenger model asynchronously, comparing its output against the production model without affecting the user's experience.

Optimization Architect

Name: Optimization Architect
Author: coreymaypray

bycoreymaypray

0•

Analytics & Monitoring

Implements autonomous cost governance, multi-model routing, and performance monitoring for Anthropic AI applications.

Optimization Architect is a specialized Claude Code skill designed to manage and optimize AI operational costs within the Maycrest ecosystem. It transforms Claude into a ruthless financial governor that builds robust routing logic between Claude Haiku, Sonnet, and Opus models while implementing hard circuit breakers to prevent runaway API spend. By leveraging Supabase Edge Functions and Postgres for telemetry, it enables shadow testing and LLM-as-a-Judge evaluations to ensure that cheaper model migrations only occur when quality is statistically proven, providing a mathematical guarantee of system stability and cost-efficiency.

Key Features

010 GitHub stars

02Multi-model routing logic for Claude Haiku, Sonnet, and Opus

03LLM-as-a-Judge scoring systems for quality-controlled model promotion

04Shadow testing infrastructure for risk-free model evaluation

05Autonomous circuit breakers for cost and error threshold protection

06Postgres-based cost telemetry and token usage dashboarding

Use Cases

01Protecting production systems from token drain attacks or infinite loops

02A/B testing new model versions against live traffic using shadow execution

03Reducing Anthropic API costs by routing simple tasks to Claude Haiku

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add coreymaypray/sloth-skill-tree optimization-architect

For use in Claude.ai and ChatGPT

Key Features

010 GitHub stars

02Multi-model routing logic for Claude Haiku, Sonnet, and Opus

03LLM-as-a-Judge scoring systems for quality-controlled model promotion

04Shadow testing infrastructure for risk-free model evaluation

05Autonomous circuit breakers for cost and error threshold protection

06Postgres-based cost telemetry and token usage dashboarding

Use Cases

01Protecting production systems from token drain attacks or infinite loops

02A/B testing new model versions against live traffic using shadow execution

03Reducing Anthropic API costs by routing simple tasks to Claude Haiku

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add coreymaypray/sloth-skill-tree optimization-architect

For use in Claude.ai and ChatGPT