HTML Render Check

The render check is the rendering layer of HTML verification. Static checks (token compliance, nav chain, prose) read the file; this skill loads it in a real Chromium engine (Playwright + chromium-headless-shell) and asserts on what actually rendered: geometry, visibility, runtime errors, network, and pixels. A finding is an adjudication prompt, not authorization to edit — some overlaps are intentional design (badges, callouts, straddling markers).

Position in the check stack

This skill owns two layers of HTML verification and runs before the artifact reaches the learner/user:

1. Structure layer (offline, parse-level; absorbed from the former html-review skill): DOCTYPE, head/body/title/charset, duplicate ids, tag closure (lxml cross-validation), external-resource inventory, inline event handlers. scripts/validate_html.py; --strict adds html5lib spec parsing + VNU W3C checker (optional deps).
2. Render layer (real Chromium): layout geometry, JS runtime errors, broken subresources, screenshot evidence — what parsing cannot see.

Between them sit the target skill's own static checks:

• teach: css-self-check / nav-chain-check / prose / beat checks run between the two layers; check.sh runs structure → render around them.
• impeccable: in the finish/inspect pass, when no real browser (chromium/chrome) is on PATH — this skill brings its own browser. It complements impeccable detect (which reads the file) with actual render evidence.

Structure expects complete documents (teach lessons, surface artifacts) — not fragments.

Setup