Cyclable Auto-Approve Profile for --yolo Mode

Mistral VibeSep 4, 2026View original changelog

Mistral Vibe changed how its widely used --yolo flag behaves: it now selects a named auto-approve agent profile rather than a one-off permission bypass. Because it is a profile, users can cycle into and out of it mid-session with Shift+Tab, and administrators can block it entirely via disabled_agents. The change turns a blunt "skip all confirmations" switch into a documented, toggleable, and governable mode.

Key Takeaways

Mistral Vibe's --yolo flag no longer flips an isolated bypass; it now selects a named auto-approve agent profile.
Because it is a profile, users can cycle in and out of auto-approve mid-session with Shift+Tab instead of relaunching.
Organizations can block the auto-approve profile outright via the disabled_agents config option.
The change pairs with a related fix ensuring the mode indicator stays accurate across --yolo, --auto-approve, and bypass_tool_permissions.
--yolo is one of Mistral Vibe's most-referenced flags in community setups, so behavior changes here reach a wide audience.
The move reflects a broader pattern in this release of making permission and approval state more explicit and administrable, not just faster.

From a Flag to a Profile

--yolo has long been Mistral Vibe's shorthand for "approve every tool call without asking," the equivalent of --auto-approve. In v2.25.0, Mistral Vibe changed the underlying mechanism: passing --yolo at launch now selects a distinct auto-approve agent profile, one of the named modes the CLI's permission system already understands, rather than flipping an isolated boolean.

Mid-Session Cycling and Visibility

Because auto-approve is now a proper profile, it slots into the same Shift+Tab cycling gesture Mistral Vibe already uses to switch between agent modes mid-session. A user who launches in --yolo mode is no longer locked into it for the whole session; they can cycle back to a confirmation-based mode without restarting, and just as easily cycle back into auto-approve later. This is paired elsewhere in the release with a fix that keeps the mode indicator honest about the effective tool-approval bypass across --yolo, --auto-approve, and bypass_tool_permissions, so the UI now reflects reality regardless of which mechanism enabled it.

Administrators Can Turn It Off

The profile model also makes auto-approve governable for the first time: it can be blocked via disabled_agents, letting teams and organizations disable the fully-permissive mode outright rather than relying on individual users never reaching for the flag. For a tool that runs arbitrary shell commands and file edits on a user's behalf, making the highest-risk mode an explicit, nameable, and administratively blockable profile is a real improvement to how the permission system is reasoned about, even though it ships as a "Changed" item rather than a new feature.