QA: manual E2E testing over a branch

Driving Kiln's real UI, as a user would, to find what's broken, not code review, not the automated Playwright suite. Read the playwright skill's SKILL.md and references/driving_the_ui.md before doing anything here: this skill is the process wrapper around it, and doesn't repeat its command-level mechanics or locator gotchas.

This skill finds and documents issues. It does not fix them. Note exceptions and hand back a findings report, same posture as kiln-prerelease-check.

When to use this vs. something else

• Automated regression suite (npm run tests:e2e), a separate, already-existing thing this skill has nothing to do with running or maintaining.
• The playwright skill, describes the tooling required for this skill (dev sandbox, browser commands, screenshots). Read it as part of doing QA here; it also stands alone for a coding agent that just wants to check its own fix rendered correctly. That skill is the tooling; this skill is the pass that uses it.
• This skill, a virtual QA team mate, making a pass over a bigger feature set, not one fix: real UI, end-to-end, looking for bugs, string issues, and inconsistencies across a set of screens or flows big enough that one person clicking through it alone would eat a day. Reach for it on an "E2E test", "QA pass", "manual test", or "bug bash" request.
• Line-by-line review of a diff, code-review, for reading code rather than clicking through the running app.

Process

1. Ask scope, then get oriented