Use Agent Approve

Policies and default behavior

Choose what Agent Approve allows, denies, or sends to you for a decision.

A policy controls which agent actions are allowed, denied, or sent to you for review. It combines Deny, Allow, and Ask Lists, optional Policy Rules, a default behavior, and an approval timeout. One policy is active across your connected agents at a time.

Open the Policy tab to see the active policy. Under Lists and Rules, open the Deny List, Allow List, Ask List, or Policy Rules to edit them. Tap the policy name to switch policies. Default Behavior controls actions that no list or rule has decided; On Approval Timeout controls what happens when an ordinary approval request receives no answer.

Policy Rules

Policy Rules let you describe conditions in plain language, such as Sends private customer information to a public discussion. Choose Deny, Ask, or Allow for a match. You can add rules from the library, write your own, or use Generate New Rules to draft them from a description.

The lists check command text and tool identities first. Enabled Policy Rules then use your rule evaluator to check actions that did not match a list. For example, you can block a known command in the Deny List and use a Policy Rule to ask before an agent sends private information through different tools. A list match takes precedence, so a broad Allow List entry can approve an action before any Policy Rule checks it.

Policy Rules are available on eligible Solo and Pro plans. Agent Approve Fast is powered by Jev, with SubQ Snap and Perplexity Decider through OpenRouter as included fallbacks, and uses AI credits. You can also connect your own provider account. See Policy Rules for setup, examples, thresholds, and provider settings.

Rules belong to the policy you add them to. Switching policies changes which saved rules apply. Allow All and Deny All skip both list and rule checks.

Built-in policies

Agent Approve includes policies for common starting points. You can use one as shipped or duplicate it before making changes.

PolicyBuilt-in listsDefault behavior
Restrictive AskA small Deny List of exact catastrophic commands, a broad risky-command Ask List, and an empty Allow ListApproves automatically
DefaultThe Permissive Allow List plus a curated Deny List of clearly destructive commandsAsks you
PermissiveCommon low-risk commands in its Allow ListAsks you
Restrictive BlockThe Permissive Allow List plus more than 300 destructive patterns in its Deny ListAsks you
GuardedThe same Deny List as Restrictive BlockApproves automatically
Allow AllNo lists are checkedApproves automatically
Deny AllNo lists are checkedDenies automatically
Allow All Except Deny ListDeny ListApproves automatically
Deny All Except Allow ListAllow ListDenies automatically

Allow All turns off both list and Policy Rule checks for that policy. Use Allow All Except Deny List when you want unmatched work to continue while keeping specific blocks in place.

Guarded starts with the full Restrictive Block Deny List and an Allow by Default behavior. If you add Policy Rules, they check actions left undecided by the Deny, Allow, and Ask Lists before automatic approval. Review both before relying on unattended work.

New accounts start on Restrictive Ask. Known risky-command matches ask for your approval, a small set of exact catastrophic command forms is blocked, and unmatched work is allowed. AI Policy Rules start off and can be enabled separately. These are text patterns, not a guarantee that every dangerous action is recognized. Existing accounts keep their selected policy and edits. Use Get Policy Updates to add Restrictive Ask, then select it if you want this behavior.

Default behavior

The default applies when the active lists and enabled Policy Rules have not decided an action.

OptionWhat happens to an undecided actionLists used
Ask UserAgent Approve sends the request to you.Deny, Allow, and Ask Lists
Allow by DefaultAgent Approve approves the request automatically.Deny, Allow, and Ask Lists
Deny by DefaultAgent Approve denies the request automatically.Deny, Allow, and Ask Lists

Allow by Default approves anything left undecided by the lists and Policy Rules. Review them before activating it. Deny by Default is safer for a narrow workflow, but an incomplete Allow List can stop expected work.

On approval timeout

Each policy also chooses what happens when a request sent to you receives no decision:

  • Deny is the default. The request is denied when the wait ends.
  • Auto-approve after veto window approves an ordinary request after a veto window of 30 seconds, 1 minute, 2 minutes, or 5 minutes passes with no decision.

The On Approval Timeout row on the Policy page shows the active policy's current choice. See Timeouts and offline fallback for the veto window's safety rules and limits.

Create a policy

  1. Open Policy.
  2. Open the policy selector.
  3. Tap the plus button.
  4. Enter a name and optional description.
  5. Choose Ask User, Allow by Default, or Deny by Default.
  6. Tap Create, then select the policy to activate it.

The default behavior is chosen when a policy is created. To change it later, edit the policy and choose a different option. Duplicating a policy is no longer required for a behavior change.

Policy page actions

Open the more menu on the Policy page for actions that apply to the policy collection.

ActionWhat it does
Create New PolicyOpens the policy editor so you can choose a name, description, and default behavior.
Get Policy UpdatesAdds missing built-in policies and patterns, and renames the old Restrictive policy to Restrictive Block. Existing entries and the active selection are preserved; custom policies are not touched.
Reset All to DefaultsRemoves custom policies, restores the system lists, and selects Restrictive Ask after confirmation. This cannot be undone.

Use Get Policy Updates to pick up new built-in list patterns without losing your own list entries. This does not update saved Policy Rules; review those updates from each rule’s screen. One limitation: a built-in pattern you deleted from a system default policy looks the same as a new template entry, so the update adds it back. Keep durable removals in a duplicated custom policy. Use Reset All to Defaults only when you want to discard your customizations entirely, and record any custom rules you need before resetting.

Duplicate Restrictive Block and allow ordinary work

The built-in Guarded policy already combines the Restrictive Block Deny List with automatic approval for everything else, so select it if that is all you need. Build your own copy when you also want to change the lists.

The built-in Restrictive Block policy starts with more than 300 deny patterns and asks you about unmatched commands. To keep its Deny List while allowing unmatched commands:

  1. Open Policy, then open the policy selector.
  2. Press and hold Restrictive Block.
  3. Tap Duplicate.
  4. Give the copy a recognizable name.
  5. Choose Allow by Default.
  6. Tap Duplicate, then select the new policy.
  7. Review the Deny List before relying on it for unattended work.

This setup keeps the copied Deny List active. Any enabled Policy Rules check the remaining actions before automatic approval. Review the copy before using it unattended.

Edit lists

Open Policy, choose the relevant list under Lists and Rules, then choose a tool or Global.

  • Use the plus button to add a pattern.
  • Swipe an entry to remove it.
  • Choose Exact match, Starts with, or Contains when adding a pattern.
  • Use tool-wide allow or deny controls only when every action from that tool should receive the same decision.
  • Open the more menu and choose Expand all or Collapse all to change every group at once.

Ordinary command matching ignores capitalization and collapses repeated spaces or tabs. Exact match checks the whole command, Starts with checks its beginning, and Contains checks for the text anywhere in the command. Test a new pattern with a harmless command before leaving the agent unattended.

For native exec input that is JavaScript or could also be JavaScript, an Allow entry must match the complete literal source, including case and whitespace. Prefixes and whole-tool Allow entries do not authorize that input. Some ordinary shell commands, such as ls -la, also parse as JavaScript; use the exact-source Remember option for those requests. Deny and Ask checks still apply, and input with no match still uses the policy default.

Native tools, MCP servers, and Global

The list browser groups observed tools by source:

  • Native tools are built into the coding agent, such as shell, read, write, or edit tools.
  • MCP tools belong to an MCP server. The server and tool identities are kept together so two servers can expose tools with similar names without sharing a rule accidentally.
  • Global is a cross-tool pattern bucket. Global patterns can be checked against command text and the textual or JSON representation of structured tool input. Deny checks also inspect supported MCP server and tool identity. Use Global sparingly because input formatting differs between agents and tools.

Tap the plus button to select an observed tool, search the catalog, choose Global, or add a custom tool identity. For an MCP tool, open the server group and then the tool. You can create a broad server rule or add patterns to an individual tool.

Broad rules and conflicts

A native tool can show Allow All Inputs, Deny All Inputs, or Ask for All Inputs, depending on the list. A server-scoped MCP entry shows Allow All Tools or Deny All Tools. When the broad switch is on, individual patterns in that entry are not needed.

Agent Approve prevents a broad allow rule when conflicting deny rules already exist for the same scope, and prevents the inverse conflict as well. Remove or narrow the opposite rule first. Adding an individual pattern to a broadly enabled entry turns off the broad switch so the pattern can control matching.

This guard prevents an Allow List and Deny List from both claiming the same whole tool or server. Narrow patterns can still overlap; when they do, a deny match takes precedence.

Decision order

Every list-based policy uses this order:

  1. Deny List. A match denies the action.
  2. Allow List. A match approves the action.
  3. Ask List. A match requests your approval without spending AI credits.
  4. Policy Rules. Enabled rules check the remaining action. Deny takes precedence over Ask, then Allow.
  5. Default behavior. If no list or rule decides, the selected Ask, Allow, or Deny fallback applies.

Allow All and Deny All skip both lists and Policy Rules.

If a rule cannot be checked, your If rules cannot be checked setting applies. An Allow rule cannot approve an action when another enabled rule could not be checked. See rule evaluation and failure settings.

How compound shell commands are checked

For supported shell tools, Agent Approve checks the complete command and each command it can parse from a shell sequence.

  • A Deny List match in any part denies the whole request.
  • Every part must match the Allow List before the whole request can be approved automatically.
  • An Ask List match on a remaining part sends the request to you.
  • If some parts remain unmatched by all three lists, any enabled Policy Rules check the action before the default behavior applies.

This prevents an allowed command at the beginning of a sequence from hiding a denied command later in the same request. Dynamic scripts can still behave differently at runtime, so review the full command before approving unfamiliar work.

Manage policies

Press and hold a policy in the policy selector to see its actions. Every policy can be edited: name, description, default behavior, and timeout behavior. Every policy can be duplicated. System default policies and the active policy cannot be deleted.

Go out and grab a coffee

We'll ping you if your agents need you. Try Agent Approve free for 7 days.