REAReverse Engineer Anything
繁體中文

Home / Guides / RL environments

Build RL environments.

Turn a recorded web app session into a runnable environment for training and testing AI agents.

Watch the announcement video on X

What REA builds

An agent can practice finding an issue, creating a folder, or editing a record without running the original app. REA builds the tools the agent calls, the local data those tools use, and a basic test that checks the environment starts and responds.

  1. 01 · RecordUse a test appCapture requests, responses, and the results of changes.
  2. 02 · Build with REACreate tools and local stateExport a runnable AgentEnv folder with the recording and a test.
  3. 03 · Run with AgentEnvGive the environment to an agentAdd task goals and reward checks for evaluation or RL training.
The recording determines which operations the environment can support. Changes are supported when the recording confirms their effects.

Before you start

Set up REA with your coding agent. To record an app, have a Chrome-family browser and a running test instance where the agent may create, edit, and delete data. REA records the app you provide; you start it and prepare its test data.

To run the exported environment, install the AgentEnv framework and start Docker. REA's compatibility check uses AgentEnv version 0.9.1334 with Python 3.12.

Record and build with your agent

Tell your agent which app and operations to cover. For an app with folders, adapt this example:

Your coding agent

Use REA to build an AgentEnv environment from my test web app at http://127.0.0.1:3000. Cover listing, creating, renaming, and deleting folders. You may create and remove test folders; leave existing data unchanged. Sign in as admin using the password in APP_PASSWORD. Use the browser at /absolute/path/to/chromium and export to /absolute/path/to/my-env, which does not exist yet. Check which operations are supported and report any gaps.

Example prompt. Replace the URL, operations, account, and paths. Set APP_PASSWORD in the environment that starts REA's MCP server so the password stays out of the prompt.

Your agent inspects the app, records a session, checks the supported operations, and exports the environment. After export, continue with running the environment.

Or use a saved recording

A HAR file records a browser's network requests and responses. Export one with API response bodies from the browser's network panel, then build into a new or empty folder:

Build the environment 
npx -y rea-agents@latest environment-build ./session.har ./my-env --name my-env --json

You can also pass capture Evidence JSON saved by REA's capture-browser-scenario or HAR inspection. Evidence is the saved recording with its source information. The build reads local data and writes the folder without contacting the app. Add --dry-run to review the report without writing files.

The report lists read operations, supported changes, and operations withheld because the recording did not establish their effects. Each withheld operation has a reason and a hint about what to record next.

Run the environment

Whether your agent exported the folder or you built it from a recording, run it with AgentEnv:

Run the environment test 
agent-env run ./my-env

Use the path your agent reported if it differs. The first run builds the container image and needs access to its image and package sources.

The included smoke task is a basic test. It checks the tool count and, when a suitable read tool exists, calls it and checks its response status. Passing it shows the environment starts and responds in that setup.

Add tasks and reward checks

Use the environment with your RL framework to define goals, score results, and run training. Your coding agent can help write the tasks and checks. Start with one operation the environment supports:

Your coding agent

Use this exported environment to add an AgentEnv task that asks an agent to create a folder named Practice. Check the environment state to confirm the folder exists with that name, and use that check to score the result. Reset the environment before each run. Use the AgentEnv task and verifier formats.

Example prompt for an environment that supports folder creation. A task that changes data should check the resulting state; a retrieval task can check an answer against a recorded value.

The export includes the runnable environment and its basic test. Task-specific grading is added in the evaluation workflow. See the AgentEnv documentation for tasks, verifiers, and agent connections.

Check what works

For changes to be supported, record the results as well as the action: read an item, change it, then read the item and its list again before the next change. Repeat with different values. For deletes, record the missing-item response and the updated list.

A passing startup test, an agent completing a task, and matching the original app are separate checks. For the last one, use environment-compare with a separate recording of calls made against the app from the same starting data.

Before sharing the folder, review its retained response bodies and local paths. Custom headers, query values, and body tokens can remain. Credential-name warnings identify places to review; they do not remove those values.

Follow the full recording and validation guide for coverage reports and comparison commands. The environment reference covers direct MCP serving, state controls, and limits.

Top