> ## Documentation Index
> Fetch the complete documentation index at: https://docs.deepblocker.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# How a test runs, end to end

> The eight steps from adding an agent in the console to reading its report, and who does each one.

A test goes through the same steps every time. You do most of them in the console on the agent's page. Our attacker does the talking. Our grader does the marking.

```mermaid theme={null}
flowchart LR
  A[Add agent] --> B[Prove it is yours]
  B --> C[Set rules]
  C --> D[Sign agreement]
  D --> E[Connect]
  E --> F[Plan a test]
  F --> G[Start]
  G --> H[Report]
```

## The steps

**1. Add the agent**

On the **Agents** tab press **Add an agent**. Give it a name, pick the channel, type its address and choose how we test it. See [Add an agent](https://docs.deepblocker.ai/agent-assurance/add-agent.md).

**2. Prove it is yours**

The agent starts as **Not proven yet**. Put the code we give you on the agent's site or in the bot's About text, then press **Check now**. We send no test message before this.

**3. Set the rules**

On the **Rules** tab tick the standard rules and, on Pro and above, write your own. At least one "must never" rule is needed. See [Rules](https://docs.deepblocker.ai/agent-assurance/rules.md).

**4. Sign the testing agreement**

On the **Tests** tab your workspace owner or an admin reads the agreement, types their name and role and presses **Accept and sign**. See [The testing agreement](https://docs.deepblocker.ai/agent-assurance/agreement.md).

**5. Connect the agent**

A web agent must answer our signed request. An agent on our test tools must list them. See [Connect a web agent](https://docs.deepblocker.ai/agent-assurance/connect/web-api.md) and [Connect our test tools](https://docs.deepblocker.ai/agent-assurance/connect/sandbox-mcp.md).

**6. Plan a test**

Press **Plan a test**, pick what the attacker should try, how long each conversation may be and how often it runs. The form shows how many test conversations it will use. See [Run tests](https://docs.deepblocker.ai/agent-assurance/run-tests.md).

**7. Start it**

An owner or admin presses **Start**, types the agent name and presses **Start test**. The first message goes out within a few minutes. The test list shows "N of M messages" while it runs.

**8. Read the report**

When the test finishes, a report is written. On Essentials and Pro it is published at once. On Team and above a DeepBlocker analyst reviews it first. See [Reports](https://docs.deepblocker.ai/agent-assurance/reports.md).

## What the attacker is

Every customer test uses the same neutral attacker: a customer called Sam Taylor who needs help quickly and is getting impatient. You never write the attacker's persona. The goal you pick becomes its objective.

## What happens between messages

Each attacker message is delivered to your agent on its channel. Your agent's reply comes back to us and is stored as one line of the transcript. The attacker reads it and writes the next message. A conversation stops when the attacker succeeds, gives up, or reaches the message limit.

## Where to go next

> [card] [Add an agent](https://docs.deepblocker.ai/agent-assurance/add-agent.md)
>
> The form, the address rules and the proof.

> [card] [Run tests](https://docs.deepblocker.ai/agent-assurance/run-tests.md)
>
> Plan, start, pause and stop.

> [card] [Reports](https://docs.deepblocker.ai/agent-assurance/reports.md)
>
> Sections, bands and levels.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.