execute-browser-flow

name	execute-browser-flow
description	Use when the user asks to actually run/execute/perform a test case or flow in a real browser using Playwright MCP tools, as opposed to just designing test artifacts.

Execute Browser Flow

1. Read the relevant test case(s) from test-artifacts/test-cases/.md (or the story file directly if no test cases exist yet)

2. If the flow requires unique data (e.g. registration), generate it:

   • Fields like username, SSN: unique per run (timestamp/random)
   • Other fields: static fake data unless the test case specifies otherwise

3. Use Playwright MCP browser tools to:

   • Navigate to the URL specified in the test case (or the project's base URL from CLAUDE.md if not specified)
   • Perform each step exactly as written in the test case
   • Take a snapshot to verify the expected result

4. If the flow generates reusable data (new accounts, credentials), save it to test-data/.json, appending rather than overwriting

5. Report pass/fail per test case, matching against its "Expected Result" column

6. Always call browser_close as the final action, whether the flow succeeded or failed — this is required for the video recording to be written to disk.

   Data handling rules

• Never read all of test-data/ into context proactively. Only read the specific file needed for the current flow (e.g. users.json when running a login test).
• When appending new data, read the existing file, add the new entry, write it back — don't hold prior runs' data in conversation context longer than needed for the current step.