Cost-cutting patterns turned into 10 installable skills for coding agents
Ten free, skills package common AI-cost-saving advice into installable modules for that read SKILL.md files, such as Claude Code, Codex, and Cursor. They install with one command: npx skills add Neeeophytee/ai-cost-cutter-skills. token-receipts-audit splits usage per model by both and dollar cost, because ranking models by one metric versus the other often produces opposite results, leading people to optimize the wrong number.
route-cheap-escalate-hard sets a cheap model as the default and only allows the premium model behind a specific, stated condition, refusing any escalation rule broad enough to match . advisor-call-budget caps how many times a cheap executor model can consult an expensive advisor model, and calculates the real savings from actual usage rather than a benchmark's assumed rate. cheap-swap-guard requires defining, by category, the cases where the premium model still wins before switching to a cheaper one, and routes those cases back to the premium model.
context-diet stops the agent from re-reading the same files repeatedly, cutting wasted context. Every skill ends with a check the agent actually runs to enforce the rule.
Key points
- Ten free, skills for agents that read SKILL.md (Claude Code, Codex, Cursor, etc.)
- Install with: npx skills add Neeeophytee/ai-cost-cutter-skills
- token-receipts-audit splits usage by both tokens and dollar cost per model to avoid the wrong metric
- route-cheap-escalate-hard defaults to a cheap model and gates premium-model use behind a stated condition
- advisor-call-budget, cheap-swap-guard, and context-diet cap advisor calls, define premium-model exceptions, and cut redundant file re-reads