Test the Hummingbird theme end to end

The theme ships its own checklist at docs/qa/testing-checklist.md. That file decides what gets tested, and it is the only thing that does. This skill reads it, walks the shop in a real browser, and hands back one report saying what was checked, what was found, and what nobody looked at.

Two things make the report worth reading, and both are enforced by the tooling rather than by good intentions:

• Nothing passes by default. A checklist point is green only if a check looked at it and recorded what it looked for.
• Every green points at a file. Either the check named one, or it ran inside a step and the screenshot of that step stands for it. report.js refuses to build a report with a green that has neither, and a screenshot that could not be taken is a fault on the run rather than a silence. This is why every check belongs inside a step(): the step is what photographs it.
• A campaign can be driven to the end without a person reading the report each time. Every build writes gaps.json, the points still owed an answer, and --require-complete refuses to build at all while that list is not empty.