test-feature

Parameters

When invoked, args will be a space-separated string: <ticket> <url>

• First token: Jira ticket number, e.g. NRF2-358
• Remainder: starting URL of the feature to test, e.g. http://localhost:3000/

Prerequisites

Playwright MCP manages its own browser — no existing Chrome instance is needed.

Setup

Before testing:

1. Fetch the Jira ticket details by running the script at read-jira-ticket.sh which is in the same folder as this skill file:

   bash <path>/read-jira-ticket.sh <ticket>
   

   This uses the JIRA_USER and JIRA_TOKEN environment variables, loaded from ~/.env if not already in the environment. See docs/feature-tester.md in nrf-library for setup instructions.

   If the script fails for any reason (missing env vars, network error, non-zero exit code), stop immediately. Do not proceed to any further steps. Report the exact error to the user and ask them to fix the issue before retrying.

2. Extract every scenario from the description as a numbered checklist before touching the app.

3. Navigate to the test URL using mcp__playwright__browser_navigate. Confirm the correct page loaded by checking the page title or H1 via mcp__playwright__browser_snapshot before proceeding.

Progress reporting

You must narrate every step out loud — the user cannot see tool calls and has no other way to know you are working.

Rules — follow these exactly, no exceptions

1. Before every tool call, output one short line: what you are doing and why. Examples: "Fetching Jira ticket...", "Testing AC2: navigating to /cookies/preferences", "Clicking Accept button", "Checking dataLayer push". Never call a tool without narrating first.
2. When starting each AC, announce it: "AC3: Banner should not reappear after accepting".
3. If a tool call has not returned within ~5 seconds, immediately output "Waiting for browser — <tool name> has not returned yet...". Do not silently wait.
4. If a tool call returns an error, stop and report it immediately: "FAILED: <tool name> — <error>. Stopping this scenario.". Do not silently skip the failure or move on.
5. After completing each scenario (pass or fail), output a one-line summary: "Scenario 3 done — PASS" or "Scenario 3 done — FAIL: <reason>".
6. Never proceed to the next step without narrating it first.

Browser testing rules

Clean state between scenarios

Use mcp__playwright__browser_navigate to load a fresh page for each scenario. Playwright MCP maintains isolated storage contexts — use a new navigation to reset state between scenarios rather than manually clearing cookies.

• Never try to clear HttpOnly cookies via document.cookie — it will not work. Hapi sets cookies as HttpOnly by default.
• Never POST directly to form endpoints via fetch to reset state — CSRF protection will block it (403).

Verifying state

• Use mcp__playwright__browser_snapshot to read page structure for assertions (preferred over screenshots).
• Use mcp__playwright__browser_evaluate to inspect JS globals, e.g. window.dataLayer, window.gtag, inline <script> tags injected by the template.
• Do not take screenshots — Jira ACs are text-based and there are no design mocks to verify against.

Black-box testing

Test purely through the browser UI. Do not read source code, route files, or templates unless you need to verify something that is impossible to observe through the browser (e.g. HttpOnly cookie flags, cookie expiry). If you do read code, note it explicitly in the results table.

What cannot be browser-tested

Mark these as code-verified or not testable in dev rather than silently skipping:

• HttpOnly / Secure cookie flags — cannot be read via JS; check the server-side cookie config in code.
• Cookie expiry — set server-side; verify in code.
• Server-side metrics / dashboards — require a deployed environment.
• GA measurement ID / real analytics loading — may not be configured in dev; note as a dev environment limitation, not a bug.

Cleanup

After all scenarios are complete, close the browser using mcp__playwright__browser_close.

Output format

Return only the results table and any bugs/gaps below it. No preamble, no narration of steps taken.

Scenario	Description	Result	Notes
1	...	✅ Pass / ❌ Fail / ⚠️ Partial / 🔍 Code-verified	...

List bugs, gaps, or deferred items separately below the table.