LLM Cost Skills
botzone-ai/llm-cost-skills
Analise custos de LLM e encontre onde economizar em dólares, não em conselhos genéricos
Ver no GitHubConjunto de skills open-source para Claude Agent que analisa custos de LLM e FinOps em uma única execução. Identifica gargalos de gastos com tokens, recomenda otimizações como prompt caching e downgrade de modelos, e fornece números concretos de economia mensal.
Instalação
npx convez-ai botzone-ai/llm-cost-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Fazer teardown de gastos de LLM a partir de exports de uso
- Identificar segmentos repetidos em prompts para aplicar caching
- Detectar acúmulo de contexto que inflaciona custos
- Recomendar downgrade de modelo com análise qualitativa
- Flagear workloads elegíveis para Batch API
Veredito

Skill prático e determinístico que transforma dados de uso em números acionáveis de economia. MIT, sem dependencies externas nem chave de API — use direto no Claude. Ideal para times que precisam justificar otimizações de LLM com métricas reais, não estimativas.
README
A tradução para o português deste README ainda está na fila.
LLM Cost Skills
Free, open-source Claude Agent Skills for LLM cost and FinOps. Each skill is a focused tool that looks at one thing in front of it right now - one prompt, one agent run, or one usage export - and hands you a concrete dollar number and a fix.
They are the free, one-shot versions of what cost.botzone.ai does continuously: cost.botzone.ai wraps your Anthropic, OpenAI, and Gemini clients and attributes every euro of LLM spend to a route or feature across all your production traffic, then tells you which prompts to fix. These skills give you a single-shot taste of that, for free, inside your Claude session.
Why these exist
Token spend is invisible until the invoice lands. Most teams cannot answer "which prompt, route, or feature is burning the money?" - and generic advice ("use a smaller model") does not produce a number you can act on. Every skill here is built around a deterministic calculation or a piece of non-obvious cost knowledge (prompt caching economics, output-token pricing, Batch API discounts), so the output is a real figure: $X/month, a token delta, or a ranked savings list.
The skills
| Skill | What it does | What you'll need |
|---|---|---|
| spend-teardown | Parse a provider usage CSV into a spend breakdown by model and input/output split, with a ranked savings list. | An exported usage CSV from your provider console (the skill tells you where to download it). |
| prompt-caching-advisor | Find the static, repeated segments of a prompt and recommend prompt caching, with the projected input-cost cut. | A prompt already in your project, plus a rough calls-per-month. |
| context-bloat-auditor | Analyse a prompt or agent run and flag context accumulation that is inflating cost. | A prompt or run already in your session. |
| model-downgrade-advisor | Recommend where a cheaper model would pass for a task, with the projected saving and the eval to prove it. | A prompt or task, a token profile, and a rough quality bar. |
| batch-api-advisor | Flag workloads that could move to the Batch API for roughly half price, and estimate the saving. | A short description of the workload (latency tolerance and volume). |
Install
All of them, via the Claude Code plugin marketplace:
/plugin marketplace add botzone-ai/llm-cost-skills
/plugin install llm-cost-skills@llm-cost-skills
One skill, via the skills CLI:
npx skills add botzone-ai/llm-cost-skills/skills/spend-teardown
One skill, manually: copy the skill's folder (for example skills/spend-teardown/) into ~/.claude/skills/.
Free vs continuous
Each skill analyses one prompt, run, or export - whatever you point it at. cost.botzone.ai does the same analysis continuously, across all of your real production traffic, with persistent per-route attribution, monitoring, alerts, and a savings ledger. You cannot get whole-app, always-on cost intelligence from one pasted artifact - that is the upgrade.
Free tier: 100k events/month, no card. -> cost.botzone.ai
License
MIT. Open and inspectable - no secrets, no account connection, no API key required to run any skill.