Test a local web application
Overview
Turn a user flow or reported UI bug into reproducible browser evidence. This ContextOS adaptation references Anthropic's webapp-testing skill; see source provenance and license.
When to Use
Verify a local site, reproduce a browser-only failure, inspect responsive states, or check a frontend change in Playwright. Use the environment's supported browser tools and the project's test runner; this skill does not override tool-access restrictions or require one programming language.
Rules & Patterns
Inspect existing scripts, Playwright configuration and the target route. Define the expected result and important failure path. Use an already running local server when appropriate; otherwise start the project's documented command and wait for a bounded readiness signal. Track processes started for this task and clean them up without stopping unrelated services.
Inspect rendered elements and use stable role/name/label locators. Rely on Playwright's auto-waiting and web-first assertions for the actual state. Do not use fixed sleeps or networkidle as a universal readiness requirement: polling, analytics or streaming may remain active while the tested UI is ready.
Exercise the real interaction and assert its visible result. Check a relevant error/empty case, keyboard focus, narrow viewport or reduced-motion mode when the change affects it. Collect console/page errors and useful screenshots or traces. A screenshot confirms appearance in that capture, not the success of an unasserted interaction or complete accessibility.
Use deterministic local test data. Keep credentials and private data out of screenshots, traces and logs. Use test accounts for state-changing flows; existing authorization and environment boundaries govern any external side effect.
Code Examples
For a settings form, fill controls by label, submit by role/name and assert both the saved state and the failure message for a controlled rejected save. See references/playwright.md for a minimal test pattern and its prerequisites; no helper script is assumed to exist in this skill.
Validation Checklist
- Record command, route, browser, viewport and actual assertions.
- Report failed checks, unexpected console errors and relevant artifacts.
- Distinguish real backend coverage from a controlled mock and disclose browser or platform gaps. Close task-owned browser/server resources.
Common Mistakes
Guessing selectors; only taking screenshots; replacing assertions with sleeps; hiding console failures; testing a mock and claiming the real integration works.
Integration Notes
Reuse installed Playwright and project fixtures. Installing browser binaries or changing test configuration is an implementation choice, not an automatic effect of loading this skill. No browser compatibility matrix is implied by one run.
