Оптимизация затрат на LLM в пайплайне

★ 6.8 · general

cost-optimizer is a Claude Code skill that tracks cumulative LLM costs across DAG execution and makes real-time decisions to stay within a defined budget. When spending approaches configured thresholds, it downgrades model tiers (e.g., from Opus to Haiku), skips optional nodes, or stops the pipeline early and returns partial results with a full cost breakdown. Before each node runs, the skill estimates cost from average token counts and historical data; after execution it generates a report with concrete savings recommendations, such as swapping models or parallelizing independent nodes. It is designed for teams managing execution budgets, analyzing per-node cost breakdowns, or optimizing model routing for cost — for per-node model selection, use llm-router instead.