---
title: reeval
description: >-
Evals for coding agents. Check that agents use your SDK, skills and docs
correctly.
---
> ## 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.
Does a coding agent actually use your SDK, your skills and your docs correctly? reeval answers that with evals you write like tests.
```ts title="evals/skill.eval.ts"
group(
'house greeting',
{ compare: { baseline: plain, variants: [withHouseGreeting] } },
() => {
evaluate('greets in the house style', async (c) => {
await c.send(
'Create greet.js exporting greet(name) that greets in our house style.',
)
await expect(c.exec(greetAda)).toSucceed()
})
},
)
```
An agent does the task in a fresh sandbox, here four times without the skill and four times with it (`--runs 4`). reeval checks each try and reports what the skill changed:
```text
pass house greeting › greets in the house style
arm pass
plain 0 of 4
with-house-greeting 4 of 4 helped
check plain with-house-greeting
house greeting 0 of 4 4 of 4
expected command to succeed, exit code 1
+2 more passed
…
1 pass
```
Without the skill the agent never wrote the house greeting; with it, it did every time, so reeval calls the skill's effect `helped`. When the tries differ in tool calls or time, an effort table follows.
## How it works
```mermaid
flowchart LR
E[evals/*.eval.ts] --> P[plan]
P --> A1[attempt: control]
P --> A2[attempt: with skill]
A1 --> C[checks and judge]
A2 --> C
C --> R[record in git]
R --> S[studio]
```
| Step | What reeval does |
|---|---|
| Agent | Runs Cursor, Claude Code, Antigravity, OpenCode, Goose, pi, Codex, fx or Grok Build on your prompt, in its own sandbox |
| Checks | Runs your commands in the sandbox and asks an LLM judge yes/no questions |
| Compare | Runs the same eval with and without a skill, MCP server or docs, and reports lift and effort |
| History | Saves every run as a file you commit, diff and browse |
## See every attempt
The studio shows each run's cells, pass rates and lift, and each attempt's transcript, diff, checks and judge input.
## Next
Scaffold a project and run your first eval on a real agent.
Prove a skill, MCP server or doc makes agents better.