Answer me with HTML

Answer me with HTML is an open-source agent skill for Claude Code, Codex, Cursor, OpenCode, Pi and DeepSeek Harness. The agent writes a short Markdown draft, and a CLI bundled with the skill renders it into one HTML page with diagrams, tables, code and callouts. The project says the model writes about 1/8 of the tokens it would need to hand-write the HTML.

What It Is

The skill changes how an AI coding agent answers hard questions: instead of a long block of terminal text, it returns a readable page. The model writes only content, and the am CLI handles layout, color and drawing. Pages are saved locally and are self-contained, with no CDN links or web fonts, so they open offline.

The agent decides when a page is worth it, such as for related concepts, multi-step flows or multi-way comparisons. A one-line question still gets a one-line answer.

How the Draft Becomes a Page

The draft uses Markdown with frontmatter, panel spans and fenced components. Documented components include flow, er, sequence, tree, timeline, limits, chart, annot, kv, callout, ask, code blocks and tables. Flow charts are laid out with dagre, and sequence diagrams are spaced by label width. Code blocks can quote real lines from local files, while files outside the folder and files holding keys are refused. When a draft has an error, the CLI returns the line, the component and a correct example so the agent can fix it.

Pages ship with three themes (blueprint, shadcn and paper) in light and dark modes. Each page embeds its Markdown source, and a Reply button collects answers to ask decisions and comments into one message to paste back to the agent.

Writing Check and Explainer Videos

Every render checks drafts against rules adapted from ASD-STE100, covering sentence length, plain words and passive voice. It only warns by default, with a strict mode available. am video turns a draft with one line of narration per beat into a player page, and the camera follows the node named in brackets. The README states MP4 export needs ffmpeg, and WebM export is encoded by the page itself.

Setup Path

Node.js 20 or newer is required, with no npm install step because the CLI is bundled. Install options include letting the agent follow INSTALL.md, a Claude Code plugin, a DeepSeek Harness plugin, npx skills add, or a manual copy into the agent's skills folder. An optional always-on mode adds a small page to every conclusion. The project is MIT licensed.