What problem does it solve? Convex bills on function calls and bandwidth, so a single read-heavy function can quietly dominate your bill as data grows. This Skill makes spend legible by reading your deployment's own usage evidence, attributing cost to the functions driving it, and projecting how each driver scales before it becomes a surprise invoice. ## Core Features & Use Cases - Cost Attribution: Ranks functions by bytes/documents read per call multiplied by call volume, using the official MCP insights, tables, and functionSpec tools. - Growth Projection: Shows the scaling curve of each cost driver — a full-table .collect() grows linearly with table size while indexed .take(n) stays flat — instead of false-precision dollar figures. - Cheapest-Fix Recommendations: Names the lowest-cost remedy per driver (index + .withIndex, .paginate/.take, aggregate components, caching) and emits cost-class findings for convex-expert/convex-advisor. - Confirm-Cost Gate: States the price and recurrence of any metered action (domain purchase, plan change) and requires explicit approval before proceeding. - Use Case: Before launch, ask which queries will get expensive at scale; the Skill flags that messages:list reads the whole table every call and that indexing it drops cost roughly 100x. ## Quick Start Ask the agent to preview what my Convex app will cost and rank the functions driving the most spend.