A cost-control practice: for each task inside a skill, deliberately select the cheapest model capable of doing that specific step well, rather than defaulting every step to the most capable (and most expensive) model available.
Apply: go through each existing skill and assign the specific model actually required for that step's complexity — the source offers a purpose-built prompt to do this pass automatically, bundled together with a max-iteration-cap setting in a single combined prompt so both cost controls are added in one sweep rather than configured separately.
Related: tiered-fallback-ai-cost-strategy covers a complementary API-level cost-management approach (fallback across providers rather than per-task model selection).