---
title: Recipes
description: >-
  Complete, copyable solutions to common eval tasks, each with its files,
  command and real output.
---
> ## 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.

Each recipe is one task end to end: a goal, the files, the command, the output and a few variations. Every eval and config on these pages was checked with `reeval run --dry-run`, and with the fake harness where it can run them.

```bash
npx reeval run --dry-run
```

## Skills and docs

<CardGroup cols={2}>
  <Card title="Prove a skill helps" icon="chart-line" href="/recipes/skill-lift">
    Run with and without a skill and gate on the lift.
  </Card>
  <Card title="Evaluate an MCP server" icon="plug" href="/recipes/mcp-server">
    Check the agent calls your tool, with the token brokered.
  </Card>
  <Card title="Check a recommendation" icon="book" href="/recipes/recommendation">
    Ask which library an agent recommends, and judge the answer.
  </Card>
  <Card title="Suites of cases" icon="layer-group" href="/recipes/eval-each">
    One eval per case with `evaluate.each`, pooled into one result.
  </Card>
</CardGroup>

## Agents and toolchains

<CardGroup cols={2}>
  <Card title="Compare agents" icon="code-compare" href="/recipes/compare-agents">
    Cursor, Claude Code and Antigravity on the same evals.
  </Card>
  <Card title="Python, Flutter and more" icon="box" href="/recipes/toolchains">
    Put the tools your task needs into the sandbox image.
  </Card>
  <Card title="Offline evals" icon="lock" href="/recipes/offline">
    Deny the network, or allow only the registries.
  </Card>
  <Card title="Start from an existing codebase" icon="code-branch" href="/recipes/existing-codebase">
    A template, an app or one package of a monorepo, pinned to a commit.
  </Card>
  <Card title="Capturing UI" icon="camera" href="/recipes/capturing-ui">
    Screenshot a port the agent left running and keep the PNG.
  </Card>
</CardGroup>

## Workflow

<CardGroup cols={2}>
  <Card title="Evals without Docker or keys" icon="flask" href="/recipes/fake-harness">
    Iterate against a scripted fake agent in about a second.
  </Card>
  <Card title="PR gate in GitHub Actions" icon="shield" href="/recipes/github-pr-gate">
    Fail a pull request when an eval regresses.
  </Card>
  <Card title="Lift of a pull request" icon="code-pull-request" href="/recipes/pr-lift">
    Run the base and head trees side by side and gate on the lift.
  </Card>
  <Card title="Nightly matrix and Pages" icon="globe" href="/recipes/nightly-pages">
    Run every agent nightly and publish the studio.
  </Card>
  <Card title="Query history from your agent" icon="plug" href="/recipes/agent-mcp">
    Point Cursor or Claude Code at `reeval mcp`.
  </Card>
</CardGroup>

## Case studies

<CardGroup cols={2}>
  <Card title="reeval evaluating reeval" icon="bug" href="/recipes/case-study-dogfood">
    Lift +0, yet 4× fewer tool calls: what a real suite showed.
  </Card>
</CardGroup>

## Next

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Your first eval, if you haven't run one yet.
  </Card>
  <Card title="Writing evals" icon="code" href="/guide/writing-evals">
    The guide behind these recipes.
  </Card>
</CardGroup>