A TesterArmy saved test is a list of plain-language steps that a QA agent runs in TesterArmy’s cloud. In e2e, each step becomes a line in a TypeScript test that runs on your machine or in your CI: an agent.act or agent.assert call, or a locator call when the step needs no judgment. A coding agent reads your projects and tests through the TesterArmy MCP server and writes the e2e tests.

Agent prompt

Migrate this project's TesterArmy saved tests to e2e, one group at a time. Docs: https://e2e.tester.army/docs/migrate/testerarmy.md

  1. Read the migration guide above. It maps TesterArmy project settings, steps, and groups to e2e and marks what e2e does not have yet.
  2. If you have no testerarmy MCP tools, add the server with npx add-mcp --global --name testerarmy https://tester.army/mcp. If Claude Code uses a custom CLAUDE_CONFIG_DIR, run claude mcp add --transport http testerarmy https://tester.army/mcp instead. Ask me to sign in and restart the session, then wait until I confirm.
  3. Call list_projects and ask me which project to migrate. Read it with get_project, list_project_environments, list_project_credentials, list_project_memories, list_project_files, list_groups, and list_tests. Use only read tools. Never create, update, delete, or run anything in TesterArmy.
  4. If the project has no e2e config (e2e.config.ts, .mts, .js, or .mjs), run npx e2e init --yes (or the pnpm or bun equivalent), with --engine mobile for a mobile project. Rewrite tests/example.e2e.ts for the app or delete it.
  5. Read the e2e skill (npx e2e guide prints it, npx e2e guide <topic> prints a topic) and follow it for every step below.
  6. If you ran init, ask me which model to use for agent steps: a ChatGPT, GitHub Copilot, OpenCode Console, or SuperGrok subscription, an API key, or a local model. Configure it with its provider package. If it needs a sign-in, give me the npx e2e login command and wait until I confirm.
  7. Port the project into the e2e config with the guide's "Port the project" table. TesterArmy never returns passwords. Name each credential in credentials, read its password from an environment variable, and ask me to set it.
  8. List the groups and their tests, and ask me which group to move first. Suggest a small one, or the one with the preparation test the others need.
  9. For each test, read it with get_test and write a .e2e.ts file under tests/, one awaited call per enabled step, with the guide's "Map each step" table. Turn a preparation or login test into a test.setup that saves a session. Where the guide says "Not yet" or "No equivalent", leave that test in TesterArmy and tell me why.
  10. After the last agent.assert of each test, add a locator assertion on the same end state when a label or test id names it.
  11. Run the new files with npx e2e run <file> and fix them until they pass. Keep the TesterArmy tests until I confirm they can go.
  12. Summarize which tests moved, which stayed in TesterArmy, and why.
Open in Cursor

For a migration that spans several sessions, the migrate-to-e2e skill records what moved in a progress file and keeps the TesterArmy tests until you retire each one.

​
Connect the TesterArmy MCP server

The coding agent reads your tests from the hosted server at https://tester.army/mcp. Add it to the coding agents on your machine:

npx add-mcp --global --name testerarmy https://tester.army/mcp

add-mcp writes Claude Code’s entry to ~/.claude.json. If you set CLAUDE_CONFIG_DIR, run claude mcp add --transport http testerarmy https://tester.army/mcp instead. Sign in when the client asks. Other ways to add the server are in TesterArmy’s MCP setup.

The migration only reads from TesterArmy. It never edits, deletes, or runs your tests there.

​
Port the project

Install e2e and configure a model with the quickstart. Agent calls need a model.
TesterArmye2e
web project urlapp: { url } on a web() target
mobile project and its uploaded buildsa mobile({ platform }) target with app: { bundleId, appPath }. You build the app, and device.installApp() installs it, see Mobile
static environments (Staging, QA)one target per environment, each with its own app.url. --target staging runs one
PR preview environmenta CI job that sets app.url to the preview URL, see Protected previews
login credentiala credentials entry. TesterArmy never returns the password or TOTP secret, so read it from an environment variable, see Signing in
credential totpSecreta secrets entry whose value is a function that computes the code. e2e ships no TOTP generator, so add a TOTP library
credential environment overridesone credentials entry, with E2E_USER_<NAME>_PASSWORD set per environment, see Override a value per run
inbox credentialNot yet. Call your own test inbox from a fixture
agentInstructionssystem on the agent, see Instructions and vocabulary
memoriescontext on the agent
project filesfiles in the repository, such as tests/fixtures/receipt.jpg
test viewport settingweb({ viewport: { width, height } }), one target per size
browserRegion, simulatorRegionNo equivalent. The browser runs where e2e run runs, or on a hosted provider
webSearchEnabledNo equivalent
accessibilityAuditEnabledNot yet
PR testingan e2e job on each pull request, with the GitHub reporter for the comment
group schedulea scheduled CI workflow

​
Rewrite one test

Each step becomes one awaited call. A login step becomes a setup test that signs in one time and saves a session:

TesterArmy (get_test)
{
  "title": "Creates a project",
  "steps": [
    { "type": "login", "title": "Log in as the admin" },
    { "type": "act", "title": "Open Projects from the sidebar and click New Project" },
    { "type": "act", "title": "Name the project Apollo and save" },
    { "type": "assert", "title": "Apollo appears in the project list" }
  ]
}
import { test } from '@e2e-dev/web';
import { expect } from 'e2e';

test('creates a project', { session: 'admin' }, async ({ app, agent, screen }) => {
  await app.open('/');
  await agent.act('open Projects from the sidebar and click New Project');
  await agent.act('name the project {name} and save', { params: { name: 'Apollo' } });
  await agent.assert('Apollo appears in the project list');
  await expect(screen.getByText('Apollo')).toBeVisible();
});

TesterArmy opens the project URL before the first step. In e2e, nothing opens on its own, so each test starts with app.open().

​
Map each step

TesterArmye2e
act stepagent.act('...'), or a locator call when the step opens a known route or acts on one named control
assert stepagent.assert('...'), then a locator assertion when a label or test id names the end state
login stepa test.setup that signs in and calls session.save(), and { session } on each test, see Signing in
login with temporaryEmailNot yet
files step, @receipt.jpg in a step, on webthe file’s repository path in the goal: agent.act('attach tests/fixtures/receipt.jpg'), or locator.setInputFiles()
files step on mobile (photo library)Not yet
screenshot stepapp.screenshot('label')
javascript stepbrowser.evaluate(). It fails only when the code throws, so check a { success: false } return with expect
microphone stepNot yet
step with disabled: trueleave it out
disabled testtest.skip
groupdescribe with tags, run with npx e2e run --tag <tag>
group runMode: sequential{ serial: true } on the group
group maxConcurrentRunsworkers in the config. It limits the whole run, not one group
group prepTestIda test.setup that saves a session, and { session } on the group
run_test, run_groupnpx e2e run <file>, npx e2e run --tag <tag>

Keep the agent for a step that reads the screen to decide, spans several screens, or names no stable control. Before you swap a step for a locator, find the control’s exact name, and check the same end state.

​
What differs on purpose

  • Tests run on your machine or in your CI, not in TesterArmy’s cloud. You choose the app URL and the device.
  • The model never sees a password. The runner types it from a handle, see Signing in.
  • An agent.act step replays its recorded actions without a model call. When the screen no longer matches, the agent takes over, see Replay cache.

​
Next

Agent steps

What the model sees, what it may do, and how a step ends.

Signing in

Sessions, credentials, and secrets the model never sees.

Was this page helpful?