replica-test builds a test plan from the flows in the recon map, with numbered happy, edge and negative cases, and automates what it can with Playwright specs that fail on console errors, server errors and accessibility violations.
What cannot be automated gets a manual pass. Bugs are written in a fixed format with severity S1 to S4, exact steps and evidence, and fixed test-first.
What You Get
- A numbered test plan such as F01-H1 and F01-E3 covering happy, edge and negative cases
- One Playwright spec per flow using role and label selectors, with console, 5xx and axe checks
- A manual checklist or browser-driven pass for email, OAuth, payments and visual glitches
- A
replica/bugs.mdof reproduced bugs with severity, steps, expected, actual and evidence - A fix loop: failing test first, fix, pass, keep the test
Common Use Cases
- Writing end-to-end tests for each flow of a new app
- Running an edge-case sweep for time zones, double submit, expired sessions and second-user data
- Gating a release on no open S1 or S2 bugs
What Makes It Unique
It carries a fixed edge-case checklist applied to every flow, uses severity definitions that tie S1 to data loss, security and payments, and reports only what was reproduced, keeping speculation in a separate to-check list. It also forbids load or fuzz testing of the original app’s servers.
Under The Hood
The skill is instructions plus a test plan template, a bug report template and an example Playwright spec. The agent runs shell commands: it installs Playwright and the axe package with npm, installs a Chromium browser, and runs the test suite against your local dev server. It writes spec files and markdown reports into your project and may drive a browser to screenshot steps. It uses Stripe test cards for declined-payment cases.