qingyuna/answer-me-with-html is a skill that produces a one-page HTML for agents answering complex questions. The model writes a short Markdown draft, and the bundled CLI turns it into a readable page. The latest release is v0.5.10. A single page can hold text, tables, code and diagrams, and the output is one HTML file that opens offline. On request it can also produce an explanatory video.
The key is that the model writes a Markdown draft instead of authoring the whole HTML, while the CLI generates the rest of the SVG, CSS and tags. In the repository's own benchmark, hand-written HTML averaged 4,893 tokens and the Markdown draft 612 tokens. The benchmark harness is included in the repository so the comparison can be checked again. Tokens are the unit a model uses to process input and output, so this result suggests that the draft-first approach can reduce the tokens needed to produce the output.
That said, the benchmark numbers are the repository's own measurements, and no independent verification is presented in FACTS. Agent skills that produce one-page HTML already exist, so what is new about this project is limited to a design that controls the output format and leaves the conversion to a CLI. The token-savings figures are therefore a basis for reference, not a guarantee that the same gap will appear in other environments.