Docs

Agent

Permissions

Control which shell commands the agent runs without asking, from the composer or Settings → Permissions.

By default the agent asks before it runs a shell command. You decide how much it can run on its own: with a quick choice in the composer, or with detailed rules in Settings → Permissions.

Choose an approval level in the composer

Click + in the composer and pick one of three levels. Each one is a shortcut for settings on the Permissions page, so it applies to every chat, not only the current one:

LevelWhat happensSame as
AskVelork asks before each command runs, except ones you allowed.Commands: Ask first
Approve for meCommands run on their own. Velork still asks for commands on the Always ask list and never runs the Never run list.Commands: Run automatically
Full accessAny command runs without asking, in this workspace.Run everything: In this workspace

Full access skips every prompt and every list below. Use it only in projects where you are comfortable with the agent running anything.

Approve a command

When the agent needs approval, its command card waits for you:

  • Run (Cmd + Enter) runs it.
  • Skip (Escape) skips it, and the agent carries on without it.

The menu to the left of the buttons sets what happens from now on:

  • Ask before every command
  • Auto-run commands (except safety & block lists)
  • Always allow similar for commands that start the same way
  • Always allow for this exact command

The ... menu on the card has the same choices and Copy command.

With Chat detail set to Minimal, the request shows in plain words instead, with See the command, Allow, and Not now.

Choices you make here are remembered. See Review remembered rules below.

Set rules in Settings

Open Settings → Permissions, or run Manage AI Permissions from the command palette and choose Open AI Permissions in Velork Settings.

Under Shell commands:

SettingOptions
CommandsAsk first asks before each command, except ones you allow. Run automatically runs commands on their own, except what the lists hold back.
Run everythingOff, Until restart, In this workspace, or Everywhere. When on, every command runs without asking, the Commands choice is disabled, and the command lists are hidden.
Autonomous runsRules for agents that run without you, such as task board runs. Same rules uses the rules above. Pre-approved only lets them run only what you already allowed.

Edit command lists

Under Command lists, choose This workspace or All workspaces, then edit the patterns:

  • Always ask asks even when commands run automatically. It starts with risky commands such as rm *, git reset --hard*, git push --force*, git clean -f*, and chmod *. Click Restore defaults to bring them back.
  • Never run blocks commands, even if you approve them.
  • Run without asking lets commands through without a prompt. It shows when Commands is set to Ask first.

Patterns match the start of a command, for example npm test or git *. Click Add pattern to add one. For finer rules, edit velork.bash.rules in your settings JSON.

Review remembered rules

Remembered rules lists the choices you made from chat prompts. Switch between Workspace and Session, then Remove a rule or Clear all.

You can also run Manage AI Permissions from the command palette to view or clear workspace and session rules.

MCP tools

Tools from MCP servers are managed in Settings → Tools & MCP. See MCP servers.

Next steps

Chat

The + menu and composer toggles

Tasks

Agents that run from the task board

Settings

All Velork settings

Still stuck?

Ask us anything, or see what changed in the latest release.

Changelog