Cursor Launches /orchestrate Skill to Recursively Spawn and Verify AI Agents

May 7, 2026 · Updated May 15, 2026

Cursor, the AI-first code editor built for pair-programming with AI, released the /orchestrate skill for its Cursor TypeScript SDK. This allows an AI agent to recursively spawn sub-agents—specialized agents for specific sub-tasks—to tackle ambitious goals. A planner decomposes tasks into smaller units for execution.

Token reduction: 20 percent
Cold start improvement: 80 percent
Architecture: Planner-Worker-Verifier loop
Availability: Cursor Marketplace plugin
SDK compatibility: Cursor TypeScript SDK

This update shifts the Cursor agent harness engineering focus to recursive reliability. While Cursor parallel subagents enable fast multi-file edits, this update adds automated verification. By introducing a verifier role that runs code and triggers fixes, Cursor creates a closed-loop system that improves evaluations while reducing token consumption by 20 percent.

You can now download the /orchestrate plugin from the marketplace for the Cursor SDK. This enables you to programmatically invoke autonomous loops for tasks like codebase research, which cut internal backend cold start times by 80 percent. The skill is available as a marketplace plugin for developers building on Cursor infrastructure.

View the full update on cursor.com

Cursor

@cursor_aiMay 7

Introducing /orchestrate, a skill that recursively spawns agents to tackle your most ambitious tasks with the Cursor SDK. We’ve used it to: - Autoresearch our internal skills, cutting token use by 20% while improving evals - Cut cold start times on our internal backend by 80% https://t.co/8Hm8S9J5pg

2342.8k

View on X

Still wondering? A few quick answers below.

The /orchestrate skill is a tool for the Cursor SDK that allows AI agents to recursively spawn sub-agents to complete complex engineering tasks. It moves beyond simple task execution by creating a multi-agent system where different agents handle planning, coding, and verification to ensure high-quality results for ambitious software development projects.

The system uses a Planner-Worker-Verifier architecture. A planner agent breaks down a high-level goal into smaller tasks and spawns worker agents to write the code. A verifier agent then runs the code to check for errors. If verification fails, the planner spawns a new worker to fix the issue, creating a self-correcting recursive loop.

Cursor reported that using the /orchestrate skill internally reduced token consumption by 20 percent while simultaneously improving evaluation scores. Additionally, the team used the skill to optimize their internal backend, which resulted in an 80 percent reduction in cold start times. These metrics demonstrate significant efficiency gains in both cost and system performance.

You can access the /orchestrate skill by downloading the plugin from the Cursor marketplace. It is designed to work with the Cursor SDK, allowing developers to programmatically invoke these recursive agent loops within their own workflows. The plugin is currently available for users who want to integrate advanced multi-agent orchestration into their development environment.

While standard agents often perform tasks in parallel or sequence, /orchestrate introduces a recursive verification layer. The key addition is the verifier role, which ensures that code is not just written but also tested and corrected autonomously. This builds on Cursor's existing multi-agent research to provide a more reliable and self-sufficient autonomous coding experience.