Can I prevent overspending with this skill?

Yes, the skill includes a CostTracker that monitors cumulative spend and can be configured to raise a BudgetExceededError before a call is even made if the limit is reached.

What happens if an API call fails?

The pipeline includes narrow retry logic that only attempts to recover from transient errors like rate limits or server issues, while failing fast on authentication or validation errors to save time and budget.

Why use immutable dataclasses for cost tracking?

Immutable state prevents side effects and makes auditing spend much easier in asynchronous or complex agentic workflows, as each step returns a clear snapshot of current usage.

How does model routing reduce LLM costs?

Model routing automatically directs simple, short tasks to cheaper models like Claude Haiku while reserving expensive, high-reasoning models like Sonnet for complex tasks based on input size thresholds.

Does it support Claude's prompt caching?

Yes, it provides implementation patterns for using 'ephemeral' cache controls on system prompts, which significantly reduces costs for repetitive instructions over 1024 tokens.

Cost-Aware LLM Pipeline

Name: Cost-Aware LLM Pipeline
Author: oabdelmaksoud

byoabdelmaksoud

0•

API Development

Optimizes LLM API expenses through intelligent model routing, immutable budget tracking, and efficient prompt caching patterns.

The Cost-Aware LLM Pipeline skill provides a standardized framework for managing LLM API costs in production environments without sacrificing quality. It implements four core optimization patterns: dynamic model routing based on task complexity, immutable budget tracking using frozen dataclasses, selective retry logic for transient errors, and native prompt caching for large system instructions. This skill is essential for developers building high-volume AI agents or batch processing workflows who need granular control over their API spend and observability into their token usage across different model tiers.

Key Features

010 GitHub stars

02Narrow retry logic that ignores permanent failures to save wasted tokens

03Immutable budget tracking with frozen dataclasses for reliable spend auditing

04Native Claude prompt caching patterns for repetitive system instructions

05Pre-defined pricing constants for modern model families including Haiku and Sonnet

06Dynamic model routing based on character length and complexity thresholds

Use Cases

01High-volume batch processing where token costs need to be minimized across variable task sizes

02Multi-model architectures that need intelligent logic to switch between high-reasoning and high-speed models

03Production AI agents requiring strict budget guardrails and early-exit spending limits

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add oabdelmaksoud/agention cost-aware-llm-pipeline

For use in Claude.ai and ChatGPT

Key Features

010 GitHub stars

02Narrow retry logic that ignores permanent failures to save wasted tokens

03Immutable budget tracking with frozen dataclasses for reliable spend auditing

04Native Claude prompt caching patterns for repetitive system instructions

05Pre-defined pricing constants for modern model families including Haiku and Sonnet

06Dynamic model routing based on character length and complexity thresholds

Use Cases

01High-volume batch processing where token costs need to be minimized across variable task sizes

02Multi-model architectures that need intelligent logic to switch between high-reasoning and high-speed models

03Production AI agents requiring strict budget guardrails and early-exit spending limits