> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getunbound.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Try It
> Create a policy from a prepared scenario, then exercise it from your own agent
| | |
| ------------ | -------------------------- |
| **URL** | `/try-it` |
| **Sidebar** | Try It |
| **Roles** | Admin · Analytics Viewer |
| **Requires** | A per-organization feature |
Try It turns a prepared scenario into a real policy in one click, so you can put a rule in place and then watch it fire from your own agent.
**The page has two faces.** Most organizations get the scenario list below. Where the **policy harness** is enabled, an Admin gets that instead — a browser agent with its own sandbox rules, described at the end of this page.
Creating a scenario creates a **real policy in your organization**, carrying a name that starts *Try It —*. It is enabled and scoped to no user group, so it applies to every developer from that moment, exactly like one you wrote yourself. Deleting the row deletes that policy again.
## What's on the page
| Column | Contents |
| ------------------ | ------------------------------------------------------------------------------------------------------ |
| **Vulnerability** | The weakness the scenario goes after |
| **Category** | The kind of scenario |
| **Attack Surface** | What it exercises — a terminal command or a prompt |
| **Scenario** | Something an AI agent might try |
| **Status** | **Not created**, **Awaiting test** once the policy exists, or **Tested** once it has matched something |
Two further columns carry no header: one holds this row's button, and the chevron at the end expands the row into **Why it matters**, **Try this prompt** — naming the policy it creates and its **Block** action — and, once created, **Incidents**.
A counter above the list shows how many scenarios you have created, and how many of those you have tested.
## Filters and controls
There are none. The scenarios are a fixed list, shown in full.
## What you can do
| Action | Result |
| ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Create Policy** | Creates the scenario's policy in your organization and reveals the prompt to test it with |
| Run the prompt in your agent | Exercises the new policy for real, and marks the scenario tested |
| **Delete** | Asks you to confirm with **Delete policy**, then removes that policy from your organization. Anything it already matched is kept in your logs, and the scenario returns to *Not created* |
An Analytics Viewer reads the table and the expanded detail, but gets no **Create Policy** or **Delete** button. Every scenario's policy is created with the **Block** action.
Create one scenario, run its prompt in your own agent, and you have seen a policy work end to end — authored, enforced and recorded — before you write any of your own.
## The policy harness
Where your organization has it enabled, an Admin opening **Try It** gets a coding agent running in the browser instead of the scenario list. You write a rule, give the agent a prompt that would trip it, and watch what happens — without installing anything on a machine.
The policies here are the harness's own. They are separate from your organization's policies and expire two hours after you last use them, so a harness result shows how a rule of that shape behaves, not what your live policies currently do. If you come back to an empty Policies list, they expired; set them up again.
The **Policies** control holds those sandbox rules, so a result can be read against the rule that produced it. **Skills** beside it lists the skills the sandboxed agent can reach. MCP servers are not a control on the page — you add them from inside the terminal with `/mcp add`.
| Action | Result |
| ----------------- | ----------------------------------------------------- |
| Prompt the agent | Shows how the sandbox rules treat what it tries |
| Open **Policies** | Lists the sandbox rules, and lets you add or edit one |
## Messages you may see
| Message | What it means |
| ---------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- |
| `Failed to load policy configuration. Please reload the page and try again.` | The page could not read your current policy setup, so scenarios have nothing to evaluate against |
| `Failed to create policy.` | A policy created from a scenario was not saved |
| `Failed to delete policy.` | The policy is still in place |
| `Couldn't load policy status. The Status column may be out of date.` | The list rendered, but the **Status** column may not reflect what is actually created |
| `The "" guardrail is not available for this organization.` | A Data-guardrail scenario was created on an organization that does not have that guardrail enabled |
| `The policies could not be loaded.` | The harness could not read its sandbox rules, so a result would not reflect them. **Retry** sits beside it |
| `The policies could not be saved.` | A change to a sandbox rule was not stored |
## Related
| Page | For |
| --------------------------------------------------------- | ----------------------------------- |
| [Policies — Agentic Use](/dashboard/policies/agentic-use) | Authoring the policies being tested |