Agentic Coding with Cursor for Research Engineers
For Python engineers and researchers who review Cursor's multi-file Agent edits to training and eval scripts. Research code fails quietly: a leaked split, a changed metric or an unseeded draw still prints a plausible number. You work on evalkit, a small standard-library eval harness on Python 3.12 with an offline stand-in model, so every number reproduces. The course covers Tab, Cmd-K and Agent as three scopes, .cursor/rules, checkpoints, @-mentions and .cursorignore and its limits. It ends with measuring your own review's precision and recall on seeded defects, and with how agent-assisted work reads in a take-home.
ToolChecked against Cursor Agent mode (Composer) on 2026-08-17.
Rod Rivera
Professor
Desk
Every lesson is free. Lessons marked open need no account; the rest open once you register.
What You'll Learn
- Why Tab-complete instincts fail once Agent is editing your training or eval code across files
- Which of Cursor's three modes fits a research-code task, and why
- How to write .cursor/rules that hold the agent to your split, seed and metric conventions
- How to review an agent's multi-file edit to an eval script the way a lab reviewer would
- How to use checkpoints to throw out a turn that selected on test, instead of patching it by hand
- How to catch a response cache that makes two different models agree, with a test of the behavior
- How .cursorignore, Run Mode and a hash check protect a test split, and what .cursorignore does not stop
- How to measure precision and recall of review on seeded defects, with error bars
- How agent-assisted work reads in a take-home or an interview loop
Prerequisites
- Comfort reading Python you did not write
- Python 3.12 and uv, set up in lesson 4
- Cursor installed, or willing to install it in lesson 2
Syllabus
Read the rest of Agentic Coding with Cursor for Research Engineers
The first lessons need no account. Sign in free to read the rest and keep your place on any device.
