---
title: Read each turn
description: Read each turn the harness returns in one session.
---
> ## 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 `c.send()` runs one prompt on the harness and returns that turn. The first `c.send()` is in [Evaluate](/evaluations/evaluate#send-the-task). Read the `Turn` it returns before you send the next prompt.

## Read the turn

A turn's fields stay fixed after `c.send()` returns. Read the harness's final message from `turn.text`. The `turn.status` field reports `completed`, `aborted`, or `error`. The `turn.diff` field compares the workspace at the end of the turn with its state before the harness runs. Token counts for the turn are in `turn.usage`, or `null` when the harness reports none. Messages and tool calls are in `turn.transcript`.

In the `evals/user.eval.ts` file, keep the turn the first `c.send()` returns, then read the message, the changed paths, the generated tokens, the test file, and a command:

```ts title="evals/user.eval.ts"
const turn = await c.send(
  'Add a Vitest test for getUserId in src/user.test.ts.',
)
const message = turn.text
const changed = turn.diff.files.map((file) => file.path)
const generated = turn.usage?.output
const test = await turn.readFile('src/user.test.ts')
const result = await turn.exec('npx vitest run src/user.test.ts')
```

`turn.diff.files` lists each path with a status of `added`, `modified`, or `deleted`, and `turn.diff.patch` is the patch for those files. Read the generated token count from `turn.usage.output` when the harness reports usage. `turn.readFile()` reads a workspace-relative path and returns `null` when that file is missing. Both `turn.exec()` and `turn.readFile()` run in the sandbox at the time you call them. A call to `turn.exec()` returns `exitCode`, `stdout`, and `stderr`.

For more information about asserting on a turn, see [Assertions](/evaluations/grading/assertions).

## Continue over several turns

Each later `c.send()` stays in the same attempt. The later prompt continues the same harness session, so the harness still has the earlier prompts and the replies it sent. `c.turns` lists every turn so far, oldest first.

In the `evals/user.eval.ts` file, replace the callback body with two prompts and read the first turn from `c.turns`:

```ts title="evals/user.eval.ts"
// ...
const first = await c.send(
  'Add a Vitest test for getUserId in src/user.test.ts.',
)
const second = await c.send('Make src/user.ts satisfy that test.')
const earlier = c.turns[0]
```

`second.text` is only the second reply. `second.diff` includes files from the first turn because each diff compares the current workspace with its state before the harness runs. `earlier` is the same turn as `first`. `first.text` stays the first reply. For a runnable project that checks each turn, see [`with-multi-turn`](https://github.com/invertase/reeval/tree/main/examples/with-multi-turn).

## Reference

For more information about every field on a turn, see [the Turn fields](/reference/attempt-context#turn).