Record and playback with roles and environments

Re-record the same flow for different users, roles, environments and devices, convert it, and drive it from Claude Code skills.

What you'll learn

How to record one flow several times as different pool users, against different environments and device profiles, how to keep those recordings organised, how conversion produces a single data-driven feature, and which Claude Code skills package the routine.

Why re-record

A checkout looks different for a standard user, a locked-out user and an admin; it also differs on a phone. Recording each variant with a leased pool user gives you honest fixtures for every role without hand-editing logins.

One flow, many recordings

Record as a standard user on staging

bun run automax record -p demo-shop -e staging --user standard --name checkout

Record as a problem user

bun run automax record -p demo-shop -e staging --user problem --name checkout-problem

The pool leases problem_user, applies its cached storage state (capturing it first if needed), and codegen starts logged in.

Record against local

bun run automax record -p demo-shop -e local --user standard --name checkout-local

Base URL, locale and timezone come from envs/local.yaml; the spec still uses route names, so it runs on staging too.

Record on a phone profile

bun run automax record -p demo-shop -e staging --user standard --device "iPhone 15" --name checkout-mobile

The Recorder page of the web UI: project, environment, start URL, device and pool user, then the recorded spec preview — The same recording flow from the web UI. It spawns automax record with the options you pick.

Run the recordings

bun run automax run -p demo-shop -e staging -l recorded
bun run automax run -p demo-shop -e local   -l recorded
bun run automax run -p demo-shop -e staging -l recorded --device "iPhone 15"

Convert into one data-driven feature

bun run automax record convert projects/demo-shop/recorded/checkout.spec.ts projects/demo-shop/recorded/checkout-problem.spec.ts

When several recordings share a flow, the generator proposes a Scenario Outline whose examples carry the role, and uses Given I use a leased user with role "<role>" so the feature keeps working as the pool changes:

@ui @regression @pool
Feature: Checkout
  Scenario Outline: <role> completes checkout
    Given I use a leased user with role "<role>"
    And I am on the inventory page
    When I add "{{defaultProduct}}" to the cart
    And I check out with "{{firstName}}" "{{lastName}}" "{{postalCode}}"
    Then I should see the text "<outcome>"
    # title-format: <role> → <outcome>
    Examples:
      | role     | outcome                       |
      | standard | Thank you for your order!     |
      | problem  | Error: Last Name is required  |

Review the proposal, then automax proposals accept <id>.

Claude Code skills

Two reusable skills ship with the repository so an assistant can run this routine end to end:

.claude/skills/automax-record/SKILL.md — records a flow for a given project, environment, role and device, post-processes it, runs it once and opens a conversion proposal
.claude/skills/automax-run/SKILL.md — runs a project by tag, layer, browser or process, reads the summary and the run directory, and reports failures with their before/after screenshots

Invoke them in Claude Code with /automax-record and /automax-run, or let the MCP record_start and run_tests tools do the same from other clients.