[Docs index](/docs.md) / [Tool Policies](/docs/tool-policies/overview.md) / Creating a Rule

---

# Creating a Rule

This guide walks through creating a tool policy rule, from choosing what it targets to placing it in the list.

## Before you begin

- You must be a workspace admin.
- Know which tool you want to control. If you are not sure of its name, you can look it up while you type.
- Decide whether the rule should allow or block. Most workspaces keep the default at **Allow** and add rules that block.

## Steps

### 1. Open the rule editor

Open **Permissions** and select the **Tool Policies** tab. Click **New Rule** in the top right.

### 2. Choose the action

Under **Action**, select **Deny** to block matching calls or **Allow** to permit them. Leave **Enabled** switched on unless you want to save the rule without putting it into effect yet.

### 3. Choose who the rule applies to

Under **Applies to**, pick one of the following:

| Option | Who it covers |
|--------|---------------|
| Workspace | Everyone in the workspace |
| Role | All admins, or all members |
| Group | Everyone in one group |
| User | One person |
| Subagent | Calls made by one agent |

For **Role**, **Group**, **User**, and **Subagent**, a second field appears so you can pick which one. Admins are covered by workspace rules like everyone else.

### 4. Choose the target

Under **Target**, fill in **Tool pack**, **Tool**, or both. Leave both blank to target every tool.

Both fields complete as you type. Start typing and press **Tab**:

- If one tool matches, the name is filled in.
- If several match, the text extends as far as they agree and a list opens. Press **Tab** to move through the list and **Enter** to choose.
- Choosing a tool from the list fills in its **Tool pack** for you.

You can also type a pattern. Use `*` for any run of characters and `?` for a single character. For example, `delete_*` targets every tool whose name starts with `delete_`.

A line under the target tells you what it covers. While you are still typing a name, it shows how many tools contain what you have typed. If it says no tool in the workspace matches, check the spelling before you continue, because the rule would never apply.

### 5. Limit the sources (optional)

Next to **Sources**, tick the places a call can come from if you want the rule to apply only there. Leave them all unticked to cover every source.

| Source | Calls that come from |
|--------|----------------------|
| Chat | A person chatting in Assist |
| Subagent | An agent, including scheduled runs |
| MCP | An external AI client connected through an MCP server |
| Workflow | A project workflow |
| Direct | A sandcastle app or another direct call |

### 6. Add conditions (optional)

A rule with no conditions matches on the tool and the caller alone. To look at the values in the call, click **Add condition**. See [Writing conditions](writing-conditions.md) for the details.

To have another tool judge the call, click **Add classifier**. See [Using a tool as a classifier](using-a-classifier.md).

### 7. Explain the rule

- In **Reason (shown to admins)**, write why the rule exists. This appears in the rule list and in Tool History.
- In **Message returned to the model on deny**, write what the AI should be told when the call is blocked. A good message says what to do instead, such as "Deleting rows is not allowed. Ask an admin to run this change."

### 8. Test the conditions

Under **Test conditions**, enter example values in **Params JSON** and click **Run test**. The result shows whether each condition matched. If the rule targets a single tool, click **Fill from tool** to start with that tool's parameters filled in.

This test checks conditions only. To test the whole list, including who the rule applies to, use the [simulator](testing-rules-with-the-simulator.md).

### 9. Place the rule and save

At the bottom left, use **Placement** to choose where the rule goes: **Top of chain**, **Bottom of chain**, or after a specific rule. Rules are read from the top, and the first match wins, so place exceptions above the broader rules they carve out of.

Click **Create Rule**. The rule appears in the list and takes effect on the next tool call.

![The rule editor with a target chosen by tab completion](screenshots/rule-editor-target.png)

## Next steps

- [Writing conditions](writing-conditions.md)
- [Testing rules with the simulator](testing-rules-with-the-simulator.md)
- [Managing the rule list](managing-the-rule-list.md)
- [Troubleshooting](troubleshooting.md)

---

## Navigation

### In this section: Tool Policies

- [Tool Policies](/docs/tool-policies/overview.md)
- [Use Cases and Playbooks](/docs/tool-policies/use-cases.md)
- **Creating a Rule** (current)
- [Writing Conditions](/docs/tool-policies/writing-conditions.md)
- [Using a Tool as a Classifier](/docs/tool-policies/using-a-classifier.md)
- [Testing Rules with the Simulator](/docs/tool-policies/testing-rules-with-the-simulator.md)
- [Creating a Rule from a Tool Call](/docs/tool-policies/creating-a-rule-from-a-tool-call.md)
- [Managing the Rule List](/docs/tool-policies/managing-the-rule-list.md)
- [Troubleshooting](/docs/tool-policies/troubleshooting.md)

#### Playbooks

- [Playbook: Build a Query Intent Classifier](/docs/tool-policies/playbook-query-intent-classifier.md)
- [Playbook: Control Where Your Tools Can Send Data](/docs/tool-policies/playbook-outbound-request-allowlist.md)
- [Playbook: Give a Scheduled Agent Only the Access It Needs](/docs/tool-policies/playbook-scheduled-agent-guardrails.md)
- [Playbook: Put Guardrails on Warehouse Queries](/docs/tool-policies/playbook-warehouse-query-guardrails.md)

### Other sections

- [Tool Creation](/docs/tool-creation/overview.md)
- [Subagents](/docs/subagents/overview.md)
- [Agent Skills](/docs/agent-skills/overview.md)
- [Sandcastles](/docs/sandcastles/overview.md)
- [MCP Servers](/docs/mcp-servers/overview.md)
- [Scheduled Triggers](/docs/scheduled-triggers/overview.md)
- [Agent Filesystem](/docs/agent-filesystem/overview.md)
- [Workspace Permissions](/docs/workspace-permissions/overview.md)
- [Workspace Billing](/docs/workspace-billing/overview.md)
- [Chat Sharing](/docs/chat-sharing/overview.md)

[Back to docs index](/docs.md)