● Product: Cursor Agent SDK
● Agents defined in: Markdown prompts plus TypeScript functions
● Runs on: Cursor hosted, or your own infrastructure
● Included: evaluation tooling
Cursor introduced an Agent SDK that builds agents from Markdown prompts and TypeScript functions, running either on Cursor's hosted service or on your own infrastructure. Evaluation tools ship with it.
The design choice worth noting is the format. Markdown prompts and plain TypeScript functions means the agent definition lives in your repository as reviewable text rather than in a vendor's UI - which matters for version control, and matters more for the prompt-injection problem demonstrated against AGENTS.md this month.
Shipping evaluation tooling in the SDK rather than as a separate product is the other signal. The gap between agents that demo and agents that hold up is measurement, and most agent frameworks still leave that to you.
Sources
The Neuron, 27 October.