Cursor 3.8 adds /automate skill and Slack triggers for hands-off AI agents

Cursor shipped version 3.8 on June 18, pushing its Automations feature well past a simple scheduler. The headline addition is a /automate skill that lets you build an always-on agent straight from a normal chat session: you describe the job in plain language, and Cursor wires up the triggers, instructions, and tools for you. The release also adds Slack and GitHub triggers and turns on computer use for cloud agents by default. The changes are documented in Cursor's official changelog.

What's actually new

Until now, kicking off a Cursor Automation meant configuring it by hand. With /automate, you can say something like "every time a PR review requests changes, open the files and draft a fix," and the agent assembles the automation itself. Cursor confirms the skill runs from a local agent session, so the setup step now lives where you already work.

The trigger surface is wider, too. Cursor added five new GitHub triggers — issue comments, inline PR review comments, PR review submissions, review-thread status changes, and completed GitHub Actions workflow runs — alongside a new Slack trigger that fires when someone reacts to a message with a designated emoji. In other words, a teammate can flag a Slack message with an emoji and quietly hand the work to an agent, as AlternativeTo reported.

Cloud agents that can see a screen

The automations don't run on your laptop. They spin up Cursor Cloud Agents, which work in a hosted environment so the job continues even when your machine is asleep — and you can pull an agent back down to local to test its changes yourself. New in 3.8, those cloud agents have computer use enabled by default, meaning an automation can drive a browser or app to produce a demo or artifact of its work, not just edit files.

The release rounds things out with quality-of-life fixes: automations can be saved in an incomplete state without losing progress, they open pull requests by default, and they can delete memory files on request. Cursor also published marketplace templates for common jobs like GitHub Actions triage and auto-fixing PR review comments.

Why it matters

This is part of a broader 2026 shift from autocomplete toward delegation — handing whole issues, reviews, and cleanup tasks to agents that act on events rather than on a keystroke. Trigger-based automation is powerful, but it also means agents act when you're not watching, so guardrails matter more than ever: scope what each automation can touch, require review before merges, and keep secrets out of reach. For people learning to build with AI, 3.8 is a clear signal of where the tools are heading — agents that wait for an event, then do the work.