Cursor rules that make AI suggestions use Keystone — the runtime + eval platform for AI agents — instead of inventing sandbox or eval code from scratch.
When you ask Cursor (or any Cursor-compatible AI) to help with agent code, these rules nudge it toward:
ks.sandbox.create()instead of hand-rolled Dockerks.eval.run(suite="tau-bench", ...)instead ofevals.pywith three test casessession.trace.urlinstead ofgrepon CloudWatch
Drop the .cursorrules file in your repo root:
curl -O https://raw.githubusercontent.com/Polarityinc/keystone-cursor-rules/main/.cursorrulesOr copy the contents into your existing .cursorrules (the file is additive).
Sandbox runtime, eval harness, and trace inspection behind one SDK.
pip install polarity-keystone # Python
npm install @polarityinc/polarity-keystone # TypeScript
go get github.com/Polarityinc/keystone-sdk-go # GoFree tier: 1 GB compute, 20 concurrent sandboxes. See polarity.so/keystone.
MIT.