LLM Cost Optimization

llm-cost-optimization

Reduce LLM API costs without sacrificing quality. Covers prompt caching (Anthropic), local response caching, prompt compression, debouncing triggers, and cost analysis. Use when building LLM-powered features, analyzing API costs, optimizing prompts, or implementing caching strategies.

majiayu000/claude-skill-registry·0installs663starsSecurity

SKILL.md

Full skill instructions

LLM Cost Optimization

Practical techniques to reduce LLM API costs by 35-65%.

Quick Reference

Technique	Savings	When to Use	Reference
Prompt Caching	25-45%	Same system prompt, frequent calls	caching.md
Response Cache	100%	Repeated identical requests	caching.md
Prompt Compression	10-20%	Long system prompts	prompts.md
Debouncing	50%+	Duplicate triggers	triggers.md

The 80/20 of LLM Costs

For short user inputs, system prompts dominate costs:

Text Length	Input Tokens	System Prompt %
Short (~100 chars)	~250	80-87%
Medium (~500 chars)	~450	44%
Long (~2000 chars)	~900	22%

Optimization priority:

Cache system prompts (biggest impact)
Cache identical requests (free repeats)
Debounce triggers (prevent waste)
Compress prompts (last resort)

Cost Estimation (Claude Haiku 3.5)

Text Length	Est. Cost
Short (~100 chars)	~$0.0004
Medium (~500 chars)	~$0.0008
Long (~2000 chars)	~$0.002

Benchmark: 1000 translations ≈ $0.80 (before optimization)

Implementation Checklist

Before Building

Add logging to every AI trigger point
Verify triggers fire exactly once per user action
Check for Pressed/Released event duplicates

Caching Strategy

Enable Anthropic Prompt Caching for system prompts
Implement local response cache (hash-based)
Include model name in cache key
Set reasonable cache limits (e.g., 500 entries LRU)

Prompt Design

Measure current token count
Identify critical rules (security, output format)
Test quality after compression
Document WHY for each rule kept

Common Mistakes

Mistake	Impact	Fix
Trigger fires twice	2x cost	Check event.state
No prompt caching	Full price every call	Use cache_control
Aggressive prompt compression	Quality drops	Keep critical rules
Cache key missing model	Wrong results	Include model in key

Quick Wins

1. Check for Duplicate Triggers

// Before ANY optimization, verify this
log::info!("AI trigger fired: {:?}", event);
if event.state != ShortcutState::Pressed {
    return;  // Ignore Released events
}

2. Enable Prompt Caching (Anthropic)

let system = vec![SystemBlock {
    block_type: "text".to_string(),
    text: system_prompt,
    cache_control: CacheControl { cache_type: "ephemeral".to_string() },
}];

3. Add Response Cache

// Check cache before API call
if let Some(cached) = get_cached(&text, &model) {
    return Ok(cached);  // Free!
}

// Save after API call
save_to_cache(&text, &result, &model)?;

Anti-Patterns

TOON format for plain text - Only helps with structured data
Caching without model key - Haiku vs Sonnet give different results
Prompt compression first - Optimize triggers and caching before touching prompts

Install skill

Copy and run in your terminal

npx -y skills add https://github.com/majiayu000/claude-skill-registry --skill llm-cost-optimization

View on GitHub

skills/data/llm-cost-optimization/SKILL.md

More Security skills

View all

Security

supabase

Handles the full Supabase workflow from schema changes to deployment, with built-in security guardrails that catch common traps like RLS...

2.7K 308.9K

View

Security

neon-postgres

Comprehensive guides and best practices for Neon Serverless Postgres, covering setup, connection methods, authentication, and platform APIs.

98 216.2K

View

Security

firebase-auth-basics

Guide for setting up and using Firebase Authentication. Use this skill when the user's app requires user sign-in, user management, or secure data access using auth rules.

462 163.8K

View

Security

firebase-security-rules-auditor

A skill to evaluate how secure Firestore security rules are. Use this when Firestore security rules are updated to ensure that the generated rules are extremely secure and robust.

462 127.7K

View

Security

firebase-ai-logic-basics

Official skill for integrating Firebase AI Logic (Gemini API) into web applications. Covers setup, multimodal inference, structured output, and security.

462 125.1K

View

Security

better-auth-best-practices

Complete Better Auth server and client setup with database adapters, session management, plugins, and security configuration.

222 118.2K

View

Security

vercel-cli-with-tokens

Deploy and manage projects on Vercel using token-based authentication. Use when working with Vercel CLI using access tokens rather than interactive login — e.g. "deploy to vercel", "set up vercel", "add environment variables to vercel".

31.9K 116.1K

View

Security

cloudflare

Complete Cloudflare platform integration with decision trees for compute, storage, AI, networking, security, and infrastructure-as-code.

3K 110.7K

View

Security

azure-compliance

Run Azure compliance and security audits with azqr plus Key Vault expiration checks. Covers best-practice assessment, resource review, policy/compliance validation, and security posture checks. WHEN: compliance scan, security audit, BEFORE running azqr (compliance cli tool), Azure best practices, Key Vault expiration check, expired certificates, expiring secrets, orphaned resources, compliance assessment.

253 103K

View