Improvements to Cursor Automations
agents
read at source ↗ cursor.com
Improvements to Cursor Automations
Source: Cursor Date: 2026-06-18 URL: https://cursor.com/changelog/06-18-26
Summary
Cursor shipped three substantive additions to its Automations feature: a /automate skill that configures triggers, instructions, and tools from a plain-language description; five new GitHub event triggers (issue comments, PR review comments, review submissions, review thread updates, workflow completion); and a Slack emoji trigger that kicks off automations by reacting to a message. Cloud agents can now also use computer use — running a browser or desktop environment to produce demos or artifacts. Two marketplace templates were added for GitHub workflow triage and PR review automation.
Implications
- Coding-agent competition. The five new GitHub triggers significantly expand Cursor Automations’ footprint in the software development lifecycle — from code editing into CI/CD and PR review workflows that previously required GitHub Actions or custom webhooks. Combined with Slack emoji triggers, Cursor is building ambient automation hooks that activate on developer workflow events, not just explicit prompts. This is the IDE-as-orchestrator pattern consolidating.
- Agentic engineering patterns — computer use in cloud agents. Giving cloud agents a browser/desktop environment to produce demos or artifacts is a meaningful capability expansion: agents can now generate visual outputs, interact with web UIs, and verify their own work visually. The pattern (agent spins up headless environment, acts, returns artifact) is increasingly common across Cursor, Claude Code, and Codex CLI — it is becoming a standard agentic primitive.
- Agentic engineering patterns — natural-language automation configuration. The
/automateskill that infers triggers and tools from a description is an early signal that the configuration surface for automation platforms may collapse into natural language. If that pattern matures, the gap between “writing an automation” and “describing what you want” narrows to near zero for common cases.