---
title: Quick start
description: reeval shows you whether a change helped a coding agent.
---
> ## Documentation Index
> Fetch the complete documentation index at: https://docs.reeval.ai/llms.txt
> Use this file to discover all available pages before exploring further.

reeval runs a coding agent in a sandbox and checks the outcome. Here you run the same harness as a baseline without the skill and a variant with the skill, then see whether the skill helped.

To have a coding agent follow these steps, use the [setup skill](https://github.com/invertase/reeval/tree/main/skills/setup).

## Before you begin

This example uses Docker for its sandbox. You need these:

| You need | For |
|---|---|
| Node {{ node }} or later | reeval's CLI |
| Docker or Podman | the sandbox |
| An API key for the agent harness | the harness |
| A Google API key (optional) | insights, the model's explanation of each evaluation after a run |

For more information about other harnesses and their keys, see [Harnesses](/configuration/agents).

For more information about Podman or Docker on macOS or Windows, see [Sandboxes](/configuration/sandboxes).

## 1. Set up reeval

Install reeval in the project:

```bash
npm install -D {{ pkg }}
```

Create the project files for the harness:

```bash
npx reeval init --harness claude-code
```

The command writes the `reeval.config.ts` file, an example evaluation in `evals/greet.eval.ts`, and the `evals/environments.ts` file, and names the keys it uses. Add them to the `.env.local` file next to the `reeval.config.ts` file. The Google key turns insights on. A run without it skips insights and still runs the evaluation:

```bash title=".env.local"
ANTHROPIC_API_KEY=replace-with-anthropic-key
GOOGLE_GENERATIVE_AI_API_KEY=replace-with-google-key
```

Build the image the harness runs in:

```bash
npx reeval image build
```

For more information about image builds, see [Toolchains](/configuration/environments/toolchains).

Check Node, Docker, the image, and the keys with the `reeval doctor` command:

```bash
npx reeval doctor
```

The command highlights any environment issues. When it finds none, the output ends with this line:

```text
No problems.
```

Without the Google key, the output ends with this line instead, and you can continue:

```text
No problems, 1 warning.
```

## 2. Write the evaluation

The baseline is an unstyled page. Create the `evals/project/index.html` file:

```html title="evals/project/index.html"
<!doctype html>
<html lang="en">
  <head>
    <meta charset="utf-8" />
    <title>Release notes</title>
  </head>
  <body>
    <h1>Release notes</h1>
    <p>Version 2.0 is out, with faster builds and a new command line.</p>
  </body>
</html>
```

The variant adds Anthropic's `brand-guidelines` skill. Add the baseline and the variant to the end of the `evals/environments.ts` file:

```ts title="evals/environments.ts"
export const page = defineEnvironment('page', ({ directory }) => ({
  files: directory('./project'),
}))

export const withSkill = page.extendEnvironment(
  'brand-guidelines',
  ({ skill }) => ({
    skills: [
      skill.git('github:anthropics/skills', {
        ref: 'dbd4588f9e1033efb41dad4bef2f7947c8993d44',
        path: 'skills',
        only: ['brand-guidelines'],
      }),
    ],
  }),
)
```

Create the evaluation:

```ts title="evals/brand.eval.ts"
import { evaluate, expect } from '{{ pkg }}'
import { page, withSkill } from './environments.ts'

evaluate(
  "styles the page in Anthropic's brand",
  {
    environment: page,
    compare: { variants: [withSkill] },
    runs: 4,
    gate: { passRate: 0.75 },
  },
  async (c) => {
    await c.send(
      "Give index.html Anthropic's look: write styles.css with Anthropic's " +
        'brand colors and fonts, and link it from index.html.',
    )

    await expect
      .soft(c, 'links styles.css')
      .toHaveFile('index.html', /href="styles\.css"/)
    await expect.soft(c, 'brand orange').toHaveFile('styles.css', /#d97757/i)
    await expect(c, 'brand fonts').toHaveFile(
      'styles.css',
      /Poppins[\s\S]*Lora|Lora[\s\S]*Poppins/,
    )
  },
)
```

The prompt doesn't name the brand's color or fonts, so the assertions pass only when the harness knows them. The variant passes when at least 3 of its 4 attempts pass. For more information about why this evaluation uses four attempts, see [Statistics](/reference/statistics#four-attempt-minimum).

## 3. Run the comparison

Run the eight attempts:

```bash
npx reeval run brand
```

The run prints a block like this one:

```text
pass             styles the page in Anthropic's brand
  arm               pass
  page              0 of 4
  brand-guidelines  4 of 4  helped
  check             page    brand-guidelines
  brand orange      0 of 4  4 of 4
  brand fonts       0 of 4  4 of 4
  +1 more passed
```

The variant passed both brand assertions in 4 of 4 attempts and the baseline in 0 of 4, so the effect reads `helped`. The counts can differ.

## Next steps

For more information about adding variants or gates, see [Measure lift against a baseline](/results/compare/lift).