Show HN: Get your coding agent off ~2019 JavaScript, onto modern web primitives

A new Google-led project aims to steer AI coding agents away from legacy JavaScript patterns toward modern HTML, CSS, and browser APIs, promising smaller, faster, more accessible web apps. Commenters welcome the idea of encoding up‑to‑date best practices for both humans and agents, but raise concerns about Chrome’s outsized influence on standards, the push toward Chrome-first features, and the risk of agents generating complex, hard-to-maintain or less compatible code. Many emphasize progressive enhancement, cross‑browser “Baseline” support, and clear constraints on browser targets as essential safeguards if such guidance is widely adopted.

Purpose of the project

  • Repository provides “skills” to steer coding agents (and humans) toward modern HTML/CSS/JS primitives instead of legacy patterns and heavy JavaScript.
  • Focus is on UX, performance, accessibility, and smaller code, not on “pretty HTML” or developer fashion.
  • Many welcome it as a practical way to replace verbose, brittle JS (e.g., ResizeObserver/setTimeout hacks) with native features.

Best practices and AI‑generated code

  • Some argue “best practices” are fading because people won’t read AI‑generated code.
  • Others insist fundamentals still matter: cleaner code reduces context for both humans and agents, lowers token use, and reduces accidental breakage.
  • Concerns about agents accreting messy codebases, accumulating technical debt, and eventually “trapping” agents in incoherent states.
  • Skepticism that developers will actually review agent output, even if they should.

Browser compatibility and Chrome influence

  • Strong worry about reinforcing Chrome’s dominance and re‑creating a “best viewed in X browser” world.
  • Critics point out many recommended features are Chrome‑stable but missing or experimental in Firefox/Safari, sometimes with tracking/analytics overtones.
  • Project contributors counter that:
    • Firefox and Edge teams are involved.
    • Guidance is supposed to default to “Baseline Widely Available” features with progressive enhancement or fallbacks.
  • Debate over whether this is honored in practice, and whether progressive enhancement is realistic for most teams vs targeting only the cross‑browser intersection.

Distribution and agent integration

  • Multiple requests for simple HTTP/markdown access (e.g., an index or llms.txt) so agents can fetch guidance without npm or custom tooling.
  • Idea of tiny-core agents that dynamically fetch skills on demand.
  • Some report high token cost when using the skills; others argue deterministic linters could handle part of this without tokens.

Scope, effectiveness, and alternatives

  • Legacy scanner currently finds relatively few issues; maintainers say it’s tuned to minimize false positives and will expand.
  • Some developers want the opposite tool: enforce conservative, highly compatible code and avoid experimental APIs.
  • Mixed sentiment: some find the guides genuinely useful; others plan to avoid them to keep sites simple or to resist Chrome-centric evolution.