Cursor Launches /orchestrate Skill for Recursive Multi-Agent Task Handling via SDK

Details

• Cursor introduced /orchestrate, a new skill in its TypeScript SDK (@cursor/sdk) that recursively spawns agents to manage complex tasks.
• Used internally to autoresearch skills, reducing token usage by 20% and improving evaluation metrics.
• Achieved 80% faster cold start times on internal backend services.
• Builds on multi-agent research with verifiers: planners spawn workers to write code and verifiers to test it; failures trigger new workers for fixes.
• Accessible via plugin download from Cursor's marketplace; integrates with SDK features like sandboxed VMs, repo indexing, semantic search, and model routing (e.g., Composer 2 for boilerplate, Opus or GPT-5.5 for complex tasks).
• SDK enables programmatic access from any pipeline, with persistent runs, PR creation, and artifact attachments.

Impact

Cursor's /orchestrate advances agent orchestration beyond rivals like OpenAI's open-source Symphony by embedding recursive verification in a proprietary SDK, enabling production-grade CI/CD integration with proven efficiency gains (20% token savings, 80% faster cold starts). Benchmarks show Cursor SDK boosting GPT-5.5 performance to 87.2% from Codex's 61.5%, pressuring competitors to match closed-loop agent reliability. This shifts developer workflows toward supervision, accelerating adoption in enterprise devops while raising lock-in concerns versus open alternatives.