Sign InOpen Brain
CursorEngineering PostOfficial Source

How we set up our cloud agent environment

Cursor’s cloud-agent adoption grew after it treated the dev environment as agent infrastructure: Linux parity, one discoverable CLI, end-to-end testing, and automated repair.

Cursor · Jul 30, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Cursor moved its monorepo onto agent-ready Linux VMs, centralized startup and utilities in an internal CLI, and added end-to-end computer-use testing. Cloud agents went from **roughly 1 in 10 merged PRs in December** to **more than half today**.

Practical Implication

Treat the development environment as part of the agent harness. Match cloud and local dependencies, replace fragile command sequences with one discoverable interface, expose verification workflows, and let supervisors own long-running processes.

Agent-Ready Context
Cursor moved its monorepo onto agent-ready Linux VMs, centralized startup and utilities in an internal CLI, and added end-to-end computer-use testing. Cloud agents went from **roughly 1 in 10 merged PRs in December** to **more than half today**.

Treat the development environment as part of the agent harness. Match cloud and local dependencies, replace fragile command sequences with one discoverable interface, expose verification workflows, and let supervisors own long-running processes.

Skills alone only helped at the margins because the underlying commands remained convoluted. Cursor also built **Cursor Cloud MCP** and **Cloud Doctor** to diagnose and repair environment failures, but the adoption figures are internal and do not isolate which change caused the increase.
Connected Context · Feed7 Judgment

This supplies implementation detail behind Cursor’s earlier claim that complete environments drive cloud-agent reliability and adoption: dependency parity, one discoverable CLI, supervised processes, end-to-end testing, and automated diagnosis mattered more than extra instructions alone. It therefore narrows skill-centric approaches: skills cannot compensate for a fragile execution surface, and the internal adoption increase does not identify any single causal change.

What we’ve learned building cloud agentsThe new account explains the concrete environment and tooling changes behind the earlier high-level conclusion that complete development environments are central to cloud-agent reliability.Scaling Managed Agents: Decoupling the brain from the handsAnthropic separates the agent harness from replaceable sandboxes, while Cursor shows the complementary work required inside those environments to make repository tasks operable and diagnosable.Skills are new features: Building Skill-Centric Harness — Yogendra Miraje, FactSetFactSet treats skills as governed runtime features; Cursor adds the prerequisite that their underlying commands and environment must first be reliable and discoverable.addyosmani/agent-skillsThe lifecycle skill pack can encode verification workflows, but Cursor’s experience indicates such instructions add limited value when the commands they invoke remain convoluted.
Context Map
agentcoding#cloud-agents#harness-engineering#skills
Uncertainty
Skills alone only helped at the margins because the underlying commands remained convoluted. Cursor also built **Cursor Cloud MCP** and **Cloud Doctor** to diagnose and repair environment failures, but the adoption figures are internal and do not isolate which change caused the increase.