2 min lesson
What this looks like inside Cursor
Take this situation: "Your refactor agent scores 100% on your eval set, but the customer keeps hitting failures. What's the most likely problem with the eval set and how do you fix it?" Lead with your decision, then add the reason.
Step 1 of 3
What this looks like inside Cursora named benchmark and a DIY review grader
This isn’t abstract. Cursor built its own benchmark, CursorBenchCursor's internal benchmark that scores models on both task performance and token efficiency, not accuracy alone. Press Enter for the full definition., that scores models on performance and token efficiency together - reflecting the shift toward price/efficiency, not raw capability alone. On it, a top performer can sit mid-range on token spend, while a lower-effort, coding-specific variant uses far fewer tokens at slightly lower performance. The headline efficiency metric is tokens per $100 spent. That dual-axis is the same instinct you want in your own eval set: don’t just ask did it pass, ask what it cost to pass.
You can codify the LLM-judge layer as a Cursor skill: “You are an expert code reviewer. Audit this diff for X, Y, Z.” - a mini BugbotCursor's automated PR reviewer that posts inline findings and can push fix commits from isolated VMs. Press Enter for the full definition. you own.
Skill creation asks where to store it (project vs user), how to resolve conflicts, and when it auto-runs: after any feature work touching components, only when explicitly asked, or whenever tests / adjacent source files change. Wiring it to auto-run on commit turns your grader into a gate that fires on every change instead of one you have to remember to run.