11 guides
Tools, approvals and safety
What the assistant may do, and how you stay in control.
How much it may do without asking
Plan, Manual and Default modes, and the safety guard.
- Permission modes: Plan, Manual and Default, and the safety guardIn NeuralVault 2.0 you choose how much the assistant may do without asking (Plan, Manual or Default) with /plan, /manual, /default or Shift+Tab, and a safety guard with fixed rules judges every call whatever the mode.
- Local and custom models ask before every command and outbound callIn NeuralVault 2.0 a model you run yourself or reach through a custom endpoint is not trusted by default, so it asks before every shell command and every outbound call, and you can mark one trusted.
Approval cards and tool rules
Approve, decline or rewind, and see which tools need a yes.
- Approval cards: approve, decline or rewind a writeIn NeuralVault 2.0 the assistant pauses on a card before it writes, and nothing changes until you press Approve; an approved write can be rewound afterwards.
- See which tools need your approvalIn NeuralVault 2.0, Settings, Agents and tools, Tools lists every tool an agent may call, what happens if it goes wrong, and whether it asks you first or runs without asking.
- See what an agent is allowed to doSettings, Agents & tools, Tools lists every capability an agent could call, banded by the worst thing a mistaken call could do.
- Read the audit log: what an agent did and whether it was allowedIn the Beta build, the Audit logs page lists the newest 200 actions agents took on your machine, with When, Actor, Action, Target and Outcome columns; it is on the Free edition too.
Commands and folders
Give it a folder, choose PowerShell or Git Bash.
- Give the assistant a folder to work inIn NeuralVault 2.0 you attach folders the assistant may read and change; it asks before each change and cannot reach anything outside them.
- Run commands in Git Bash, and the optional Windows boundaryIn NeuralVault 2.0 the shell the assistant uses for commands is PowerShell, you can switch it to Git Bash, and you can turn on an optional boundary where Windows itself refuses writes outside your working folder; both are set in one small file, not on a Settings screen.
Agents, skills and background jobs
Pause an agent, see what runs on its own, and what skills are.
- Pause an agent so it cannot start unattended runsIn NeuralVault 2.0, Settings, Agents and tools, Pause automation has one switch per agent that stops it starting new unattended runs until you resume it.
- Background jobs, and turning one offNeuralVault runs a few scheduled jobs on their own. The Background jobs screen lists each one, when it last ran, and a switch.
- Skills: what they are and where they come fromThe Skills screen lists every Claude Code skill found on this machine. It reports what is installed; it is not a set of switches.