Cursor 3.6: Auto-review run mode for safer, longer agent runs

Cursor’s 3.6 update introduces Auto-review, a new agent run mode intended to reduce the number of approval prompts while keeping execution safer. (https://cursor.com/changelog)

What Auto-review does

Cursor says Auto-review “allows Cursor to work for longer with fewer approval prompts and safer execution.” (https://cursor.com/changelog)

Under the hood, Cursor says Auto-review applies to Shell, MCP, and Fetch tool calls. (https://cursor.com/changelog)

• Allowlisted calls run immediately. (https://cursor.com/changelog)
• Calls that can be sandboxed run in the sandbox. (https://cursor.com/changelog)
• All other agent actions are routed to a classifier subagent that decides whether to allow the call, try a different approach, or ask for approval. (https://cursor.com/changelog)

How to configure it

Cursor says you can configure the run mode in Settings → Cursor Settings → Agents → Run Mode, and you can steer the classifier agent by providing it custom instructions. (https://cursor.com/changelog)

Sources

• Cursor changelog: https://cursor.com/changelog