Add web-qa-testing skill

[spiderocious]

Summary

Adds web-qa-testing — a skill for QA-testing a running web app by driving it
in a real browser with agent-browser.

It's black-box behavioral testing: the agent exercises real user flows and
verifies what the app actually does, rather than reading or judging source code.
Because it tests through the browser, it works on any web app regardless of
stack.

This appears to be the first skill in the repo to use agent-browser, so it
also serves as a worked example of pairing the two tools.

What it covers

• Mapping a user journey before testing, then walking it with agent-browser
• The states that ship broken when only the happy path is tested: empty,
  loading, error, unauthorized, offline
• Bug reproduction (→ deterministic repro steps) and fix verification
• A PASS/FAIL/SKIP/BLOCKED result framework with bug severity
• Test plan + execution report templates

Structure

• skills/web-qa-testing/
• SKILL.md # entry point (174 lines)
• references/state-recipes.md # agent-browser recipe per state
• references/report-format.md # test plan + report templates

SKILL.md stays small; detailed recipes and templates live in references/
(progressive disclosure). Registered under a new "Testing" group in
skills.sh.json.

Verification

• skills.sh.json is valid JSON; SKILL.md frontmatter has name,
  description, license, metadata
• SKILL.md is 174 lines (< 500); reference links resolve one level deep
• All documented agent-browser commands were dry-run against a live page on
  agent-browser 0.27.1 (open, snapshot, find text/role/label, get,
  set offline, wait --state hidden)

[vercel]

@spiderocious is attempting to deploy a commit to the Vercel Team on Vercel.

A member of the Team first needs to authorize it.