Cursor Gives Cloud Agents Persistent Goals and Triggers
Cursor’s cloud agents can now react to PR and Slack events, pursue durable goals and delegate work to isolated virtual machines.
From prompted assistants to event-driven workers
Cursor released a collection of cloud-agent and harness upgrades designed to let coding agents continue operating without a person initiating every step. Cloud agents can now subscribe to pull requests, Slack conversations and schedules, waking when a relevant event occurs. An agent that creates a pull request automatically follows it, responds to continuous-integration failures and addresses automated review comments as they arrive.
The new /goal command gives an agent a persistent objective that it is expected to pursue until completion. Cursor suggests using it for outcomes such as eliminating flaky tests and restoring a passing CI pipeline. Goals can be combined with recurring check-ins and with Custom Modes, which pin a selected skill to the conversation so the agent follows the same specialized playbook throughout a long session.
Cursor also extended subagent isolation. Delegated agents may now run on separate virtual machines, each receiving a clean copy of the project and an independent working context. That permits several agents to test or investigate the same application concurrently without modifying one another’s workspace. A parent agent could, for example, dispatch a group of isolated subagents to search for different classes of defects or ask one to validate a proposed change in a fresh environment.
A steering change allows users to submit new guidance while an agent is active without abruptly terminating its current operation. The instruction is delivered at the next tool boundary, reducing the chance that intervention leaves a file operation or other action half-finished. Event subscriptions are initially limited to cloud agents.
Why it matters
The release advances coding agents beyond interactive chat toward persistent software operations. The strategically important shift is not any single command, but the combination of triggers, durable objectives, isolated delegation and mid-run steering. Together, they create the control layer needed for agents to maintain pull requests and react to development events over hours or days. That may reduce routine coordination work, but it also increases the importance of scoped permissions, observable execution and reliable stop conditions as agents gain more opportunities to act without immediate supervision.