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

# Run Cross-Agent Reviews on Agent Work

> Use recap to summarize recent agent activity, then run a cross-agent review before merging.

This tutorial shows how to use `entire recap` to understand recent agent activity, then use [`entire review`](/cli-reference/review) to run a cross-agent review before you merge.

In this tutorial, you will create a small app, make a few agent-assisted commits, recap the work, run a review with more than one agent, and use the findings to decide what to fix.

## Prerequisites

Before you start, make sure you have:

* [Entire installed](/installation)
* An AI coding agent installed
* Git configured for the project
* An Entire account
* Node.js installed

Log in to Entire:

```bash theme={null}
entire login
```

## Create a New Project

Create a Vite project:

```bash theme={null}
mkdir flight-control-review-demo
cd flight-control-review-demo
npm create vite@latest . -- --template vanilla-ts
npm install
```

Enable Entire for your agent. This example uses Codex:

```bash theme={null}
entire enable -y --agent codex
```

Replace `codex` with the agent you use.

Install another review agent so the review can compare more than one reviewer:

```bash theme={null}
entire agent add claude-code
```

This tutorial uses Codex and Claude Code as the review agents. You can use the review agents your team has installed.

Create a feature branch:

```bash theme={null}
git switch -c add-flight-control-board
```

## Build Reviewable History

Open the project in your AI coding agent and ask it to build a small app:

```text theme={null}
Build a browser app called Flight Control.

It should show a departure board for a fictional airport using local sample data.

Requirements:
- show flight number, destination, gate, departure time, status, and delay risk
- include at least 8 sample flights
- make the UI feel like an airport operations screen
- keep the data local and deterministic
- keep the code simple enough to review
```

Commit and push the work:

```bash theme={null}
git add .
git commit -m "Build flight control board"
git push -u origin add-flight-control-board
```

If Entire asks whether to link the commit to session context, choose `a`:

```text theme={null}
Link this commit to session context?
  [Y]es / [n]o / [a]lways (remember my choice): a
```

## Add More Agent Work

Ask your agent to add delay-risk logic:

```text theme={null}
Add deterministic delay-risk scoring to Flight Control.

The score should consider departure time, weather severity, gate changes, connection pressure, and current status.

Show a short reason for each score.
```

Commit and push:

```bash theme={null}
git add .
git commit -m "Add delay risk scoring"
git push
```

Ask your agent to add one more feature:

```text theme={null}
Add a "Simulate next 30 minutes" button.

Each click should apply one deterministic airport update, such as a gate change, a boarding status update, a delay increase, or a cancellation.

Keep the update sequence deterministic so the behavior is easy to review.
```

Commit and push:

```bash theme={null}
git add .
git commit -m "Add flight update simulation"
git push
```

## Recap the Work

Run `entire recap` from the project:

```bash theme={null}
entire recap
```

Use `--static` when you want plain terminal output:

```bash theme={null}
entire recap --static
```

The recap should show recent activity for this repository, including sessions, checkpoints, agent activity, and token usage when available.

[Entire recap demo](https://customer-0x5hsk8lh27fm00g.cloudflarestream.com/65415e57a6d7b2d797be3c08fe7e9dfc/iframe?muted=true\&loop=true\&autoplay=true\&poster=https%3A%2F%2Fcustomer-0x5hsk8lh27fm00g.cloudflarestream.com%2F65415e57a6d7b2d797be3c08fe7e9dfc%2Fthumbnails%2Fthumbnail.jpg%3Ftime%3D%26height%3D600)

Try a longer range:

```bash theme={null}
entire recap --week --static
```

## Configure Cross-Agent Review

Create a review profile with Codex and Claude Code as reviewers, and Codex as the judge:

```bash theme={null}
entire review --configure \
  --set-agents codex,claude-code \
  --set-judge codex \
  --set-output local
```

Entire saves the profile and tells you how to start it.

To see configured profiles:

```bash theme={null}
entire review --list
```

To see the reviewers in the selected profile:

```bash theme={null}
entire review --agents
```

## Run the Review

Run the review:

```bash theme={null}
entire review general
```

[Entire review demo](https://customer-0x5hsk8lh27fm00g.cloudflarestream.com/fc7c27edc0a1d56466182d6e811516c2/iframe?muted=true\&loop=true\&autoplay=true\&poster=https%3A%2F%2Fcustomer-0x5hsk8lh27fm00g.cloudflarestream.com%2Ffc7c27edc0a1d56466182d6e811516c2%2Fthumbnails%2Fthumbnail.jpg%3Ftime%3D%26height%3D600)

By default, Entire scopes the review against the first available mainline ref, such as `origin/main`. For stacked branches, pass the parent branch with `--base`:

```bash theme={null}
entire review general --base origin/my-parent-branch
```

To add one-off instructions for a single run, use `--prompt`:

```bash theme={null}
entire review general --prompt "Focus on risky state transitions and missing tests."
```

## Read the Verdict

When the review finishes, Entire shows each reviewer's result and then consolidates the reports into a final verdict.

Compare:

* Which issues did multiple agents find?
* Which issues were unique to one reviewer?
* Which findings are blockers before merge?
* Which findings are follow-up work?

If you want to inspect saved findings locally, run:

```bash theme={null}
entire review --findings
```

Fix the issues you agree with, then commit and push:

```bash theme={null}
git add .
git commit -m "Address review feedback"
git push
```

## Recap the Review Loop

Run recap again:

```bash theme={null}
entire recap --static
```

The recap should now include the original build work, the review run, and any follow-up fixes.

This is the core workflow:

1. Build with an agent
2. Commit and push useful chunks of work
3. Use `entire recap` to understand what happened
4. Use `entire review` to review the branch before merging
5. Fix the findings that matter

## Troubleshooting

> **`entire recap` Looks Empty**
>
> Make sure you ran agent work in an Entire-enabled repository, linked commits to session context, pushed the branch, and gave hosted analysis a few minutes to catch up.
>
> You can also check local checkpoint history:
>
> ```bash theme={null}
> entire checkpoint list
> ```

> **`entire review` Asks Me to Configure Review**
>
> Create a review profile:
>
> ```bash theme={null}
> entire review --configure --set-agents codex,claude-code --set-judge codex
> ```
>
> You can also run `entire review --configure` without flags to use the interactive setup.

> **One Reviewer Is Missing**
>
> Confirm the review agent is installed in this repository:
>
> ```bash theme={null}
> entire agent list
> entire agent add AGENT_NAME
> ```
>
> Then check the configured reviewers:
>
> ```bash theme={null}
> entire review --agents
> ```

> **The Review Uses the Wrong Base Branch**
>
> Pass the branch you want to review against:
>
> ```bash theme={null}
> entire review general --base origin/main
> ```
>
> For stacked work, use the parent feature branch instead of `origin/main`.

## Continue With

> **[review](/cli-reference/review)**
>
> See the command syntax for `entire review`.

> **[Search Past Work and Code](/learn/search-past-agent-work)**
>
> Find previous checkpoints, sessions, and commits by phrase or meaning.