v0.2.2

Roadmap

What the project intends to work on, in rough order. These are intentions, not commitments or dates: the project is maintained by one person, and a measured result can reorder everything below it. Completed work moves to the changelog.

Now: evidence for what exists

The method is argued, not measured. Closing that gap comes before adding to it.

  • A first model-evaluation run on the current release, with and without the plugin, and its results published honestly, including the cases that fail (testing).
  • A blind human comparison of pages made by 0.1.0 and the current method for the same briefs, using the protocol in testing.
  • A live-session test of the plugin: the skill loading and running end to end in Claude Code, not only installing.
  • Testing the other surfaces: a claude.ai upload and the Skills API with the release ZIP.

Next: coverage

  • Eval cases for what is not covered: a developer portfolio, an architecture studio, a large e-commerce catalogue, an immersive campaign with 3D, and a design system.
  • Windows: run the free checks and the tools on Windows and fix what breaks.
  • Templates the scan cannot read (Pug, Haml, Slim, Razor, JSP, HEEx, Go HTML): rules for the most common, so fewer projects end with incomplete coverage.
  • Headed rendering for WebGL: an option for the rendered tools to use a headed browser, for experiences that stall headless.
  • Research upkeep: a manual browser check of the Deep Beauty reference, and a periodic external link audit (research gaps).

Later: considered, not started

  • Publishing the documentation site on a schedule tied to releases.
  • A submission to Anthropic's plugin directory, when the evidence above exists. That is the owner's decision.

Not planned

  • A house style, theme or component library. The skill has no look of its own, by design.
  • Native iOS and Android guidance. The method is for the web.
  • Paid evaluations in CI, or any check that needs a secret to run on a pull request.
  • Telemetry. The tools send nothing anywhere.

Suggestions are welcome as issues. A suggestion that comes with evidence (a page the method handled badly, a measurement, a source) moves fastest.