Cursor /goal: Long-Running Objectives for Autonomous Agents
CursorCursor introduced the /goal command, which gives an agent a long-lived objective it will work toward until fully complete, rather than stopping after a single response. Users can try prompts like "/goal fix all flaky tests and make CI green" in a new chat, and pair it with a Custom Mode to follow a specific playbook or with /loop for recurring check-ins. The feature is designed for the kind of extended, unsupervised sessions Cursor's cloud agents and harness improvements are increasingly built around.
Key Takeaways
/goalturns a single prompt into a standing objective, with the agent continuing to iterate — across idle periods and multiple tool calls — until the condition is actually met.- The flagship example,
/goal fix all flaky tests and make CI green, targets a common source of agent under-delivery: stopping after a partial fix instead of driving to full resolution. /goalis explicitly designed to combine with Custom Modes, letting a long-running agent follow both a fixed objective and a fixed playbook at once.- It also pairs with
/loopfor workflows that need periodic check-ins rather than one uninterrupted run. - This is part of the same release as cloud agent Subscriptions, both aimed at the same underlying goal: agents that persist and self-direct over long sessions rather than needing constant re-prompting.
/goalreflects a broader industry trend toward goal-directed autonomous coding agents, a capability area Cursor's own community forum had been requesting comparisons to for months prior to this release.
Giving agents durable objectives
Cursor introduced /goal, a command that hands the agent a long-lived objective to pursue until it is genuinely finished, rather than treating each prompt as a one-off request. Where a normal chat message produces a bounded response, /goal sets a standing target the agent holds onto across a session — including through idle periods and multiple tool-call rounds — until the stated condition is met.
Example usage
Cursor's suggested starting point is straightforward: typing /goal fix all flaky tests and make CI green in a new chat tells the agent to keep working — investigating failures, applying fixes, re-running tests — until CI is actually green, not just until it has made one attempt.
Composability with other new features
/goal is designed to work alongside the other capabilities shipped in the same release. It can be paired with a Custom Mode to make the agent follow a specific playbook while pursuing the goal, or combined with /loop for agents that need recurring check-ins rather than one continuous run. Together with cloud agent Subscriptions, /goal rounds out a release focused on making Cursor's agents hold state and keep working over long horizons instead of needing to be re-prompted at every step.
Why it matters
Much of the friction in agentic coding tools comes from agents that stop too early — reporting partial progress and waiting for the next instruction instead of driving a task to actual completion. /goal directly targets that gap, giving users a simple way to specify a durable success condition and let the agent keep iterating autonomously until it's met.