hatch3r-browser-verify

# Skill: hatch3r-browser-verify > Last updated: 2026-07-28 ## Quick Start Invoke this skill whenever a UI-affecting change reaches a verification gate. The operating model is **spec-run-first**: write assertions once as durable Playwright specs, run them headless, and read only failures — never eyeball per-route screenshots. This converts verification cost from O(routes × screenshots) to O(failures), and specs re-run per review iteration at near-zero token cost. Two tiers plus four capabilities: - **Tier 1 (default):** scripted spec execution — functional assertions (Step 2), axe-core a11y gate (Step 3), `toHaveScreenshot()` regression diffs (Step 4), E2E scaffolds (Step 5). - **Tier 2 (exception):** token-efficient step driving via `playwright-cli` or Playwright MCP snapshot mode, only when no spec exists yet or a failure is not understood (see "Tier 2 — Exploratory Driving"). ``` Task Progress: - [ ] Step 0: Detect ambiguity (P8 B1) - [ ] Step 1: Install Playwright + axe-core (if not present) - [ ] Step 2: Spec-run verification of UI changes (Tier 1 default) - [ ] Step 3: Accessibility audit via axe-core + Playwright - [ ] Step 4: Regression screenshot diffs (diff counts, not image review) - [ ] Step 5: E2E test authoring scaffold ``` ## Step 0 — Detect Ambiguity (P8 B1) Before any browser action, scan the invocation for unresolved questions per `agents/shared/user-question-protocol.md`. Triggers: which build artifact to verify against (dev server vs `npm run build`...