richard-gyiko/which-llm ? reverse-engineered prompt

Build me a Rust command line tool that helps me pick the right LLM for a task instead of guessing.

I want it to pull in benchmark and capability data, then let me ask things like which model is best for coding, summarizing long text, tool use, or agent style workflows, while also considering cost and context size. It should be able to show the available tables, let me run SQL style queries over the data, and refresh the data from the source so it stays current. If possible, support both cached daily data and a direct API mode for fresher results.

Also make it work as an AI assistant skill, so tools like Cursor, Claude Code, Copilot, and similar can install it and automatically recommend a primary model plus a fallback based on the task. Keep the output easy to read, with a short explanation of why each model was chosen and a rough cost estimate. If you need to check the latest docs online, go ahead.