mexxi

/ittestja ​

ittestja · verb · “to test”
Tries a page or flow in a real browser and reports what is broken, without changing code.

Tests can pass while the page is still broken, because nobody looked at it in a browser.

Use it to see how a page, a flow or a pull request really behaves before you merge. You get a findings report with a severity, a screenshot and steps to reproduce for each problem; it never changes code. It needs the app to run locally and the Playwright MCP server.

Agent instructions ​

The agent follows the instructions below, from commands/ittestja.md. Its one-line description is: Run hands-on rendered QA of a route, flow, pull request, or implementation session and report findings without changing code.

Full agent instructions

Exercise the running UI named by $ARGUMENTS (a route, a flow, a pull request, or an implementation session) and return a findings report. QA observes and reports; it never edits code, commits, or switches branches.

Resolve The Target ​

  1. If $ARGUMENTS or this session identifies a retained implementation workspace, run ibni info --session "<sessionId>". For a pull request, find its workspace with ibni list by prUrl. Work in that worktreePath, start the app with its portEnv on its recorded port, and write evidence to its evidenceDir.
  2. Otherwise use the current checkout and a fresh system temporary directory for evidence. For a pull request with no retained workspace, confirm with the forge's CLI (gh pr view on GitHub, glab mr view on GitLab) that the checkout is on its head branch; if it is not, say so and stop rather than switching branches.
  3. Derive the routes, states and flows to exercise from the arguments and, for a pull request or session, from its description and diff.

Run The Pass ​

Load rendered-qa and follow it against the resolved target.

Report ​

Return the rendered-qa report: each finding's severity, what you saw, where (route, section and viewport), how to reproduce it, its evidence path, and the smallest suggested fix. Then list what was tested and worked, the checks you skipped, and the suggested executable checks.

Stop after the report. Fixes go through /wettaq's follow-up path, which adopts the pull request's workspace when this session does not own it, or a new request when there is no pull request.

Old name: /qa, which still works for one release.