/auto-review

Toggles 'Auto-review' run mode, a middle ground between allowlist and Run Everything.

/auto-review turns on Auto-review mode. A classifier checks shell, MCP, and fetch calls in order: allowlisted calls run immediately, sandboxable calls run in the sandbox, and the rest go to the classifier, which allows the call, tries a different approach, or asks for approval. Also controllable via --auto-review or in /config; allow/block rules in permissions.json steer the classifier. Existing hard approval boundaries are unchanged. Documented in the official changelog, not (yet) listed in the command reference table.

✅ WHEN TO USE IT?

Fewer prompts, without going all the way to 'Run Everything'

/auto-review — for a long session with a mix of harmless and sensitive commands

Use fixed allow/block rules per team

/auto-review — combined with matching entries in permissions.json

⛔ WHEN NOT TO?

Full manual control over every single command is wanted

Auto-review deliberately reduces the number of prompts.

Better: Leave Auto-review off, stay in the default allowlist mode

You want truly zero prompts

Auto-review still asks in unclear cases.

Better: Use /run-everything

SOURCES

Cursor Docs — Changelog ↗ cursor.com