# HTML Anything: A Zero-API-Key HTML Editor Driven by Your Existing Coding Agent

> HTML Anything is a local-first HTML editor that routes generation through whichever coding agent CLI is already on your PATH, with no additional API key required. It ships 75 skill templates across nine deliverable surfaces and exports one-click to WeChat, X, Zhihu, HTML, and PNG.

**nexu-io/html-anything** — ✨ The agentic HTML editor — your local AI agent writes the HTML, you ship it. 🚀 75 Skills × 9 Surfaces (magazine · deck · poster · XHS / tweet · prototype · data report · Hyperframes) 🛡️ Sandboxed preview · 📤 1-click to WeChat / X / Zhihu / HTML / PNG 🔑 Zero API key — Claude Code / Cursor / Codex / Gemini / Copilot / OpenCode / Qwen / Aider.

- Repository: https://github.com/nexu-io/html-anything
- Website: https://open-design.ai/html-anything/
- Stars: 8,968 · Forks: 868
- Language: HTML
- License: Apache-2.0
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/nexu-io-html-anything

## What HTML Anything Solves and Who It Is For

Engineers and writers who produce documentation, presentations, or social content through AI agents typically get markdown output. Markdown is easy to write but limited to what a renderer supports. HTML gives the author full control over layout, typography, and visual style, and it can be directly screenshotted, shared to social platforms, or converted to PNG without going through a separate layout engine.

HTML Anything is built on the premise that markdown is a draft format and HTML is what readers should receive. The README quotes a claim that Anthropic's Claude Code team stopped writing internal docs in markdown and now ships HTML. The tool's position is that since engineers already have coding agent CLIs installed and authenticated, those sessions can be redirected to generate structured HTML documents without signing up for a new service or creating additional API credentials.

The target audience is engineers and content creators who already use one of the nine supported coding agents and want to produce designed HTML output for documents, decks, posters, and social media cards. The tool is local-first and Apache-2.0 licensed.

## How HTML Anything Works: Coding Agents as the Generation Engine

HTML Anything auto-detects whichever of nine supported coding agent CLIs are present on your PATH: Claude Code, Cursor Agent, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, Qwen Coder, Aider, and IBM Bob. It then drives the selected agent to generate HTML output using a skill template as the generation context.

The mechanism described in the README is that you select a surface and a skill, provide your content as input, and the tool instructs the agent to produce a complete HTML file matching that skill's visual and structural specification. The result is a sandboxed preview within the editor. No separate API key is needed because the tool reuses the session that the coding agent CLI already has authenticated.

This is a different model from tools that call the OpenAI or Anthropic API directly. If you are not logged into Claude Code or another supported CLI, HTML Anything has no fallback. The tool is also local-first: it runs on your machine and manages files locally.

## Getting Started with HTML Anything

The project homepage is at open-design.ai/html-anything/. The README notes that a live overview page is available there before cloning. The source repository is at nexu-io/html-anything on GitHub.

The repository is a pnpm monorepo with a next/ directory for the main application and a cli/ directory for the command-line interface. The workspace package.json shows the project is organized as @html-anything/workspace and uses pnpm 10.33.2.

Each of the eight featured skills includes a real example.html file in its subdirectory that can be opened directly from the repository without authentication or setup. For example, the deck-guizang-editorial skill is in next/src/lib/templates/skills/deck-guizang-editorial/ and includes an example.html you can open locally.

The README also describes a sandboxed preview in the editor itself: generated HTML is displayed in a sandboxed preview window before export, so you can inspect the result before committing it.

## Nine Surfaces and 75 Skills

HTML Anything organizes its output capabilities into nine deliverable surfaces. The README lists these as magazine articles, keynote decks, resumes, posters, Xiaohongshu (XHS) cards, tweet cards, web prototypes, data reports, and Hyperframes videos.

The 75 skills are composable templates that define the visual and structural rules for each piece of output. The featured examples from the README include deck-guizang-editorial (a magazine-style editorial deck with ten locked layouts and five palettes including Ink, Indigo Porcelain, Forest Ink, Kraft, and Dune), deck-swiss-international (a Swiss International style deck with a 16-column grid and 22 locked layouts), doc-kami-parchment (a warm-parchment editorial document), and video-hyperframes (a Remotion-compatible storyboard with sequential 1920x1080 frames).

The Hyperframes skill is worth noting specifically: it generates six to ten sequential frames with hidden duration and transition markers and an auto-play script. The README notes these can be passed directly to heygen-com/hyperframes or Remotion to render as an .mp4 file, making the output usable in a video production pipeline, not just as a static document.

Skills are stored in SKILL.md frontmatter format in the next/src/lib/templates/skills/ directory.

## Exporting to Social Platforms and Local Files

One-click export is a central feature of HTML Anything. The README lists the export targets as WeChat (articles), X (tweets), Zhihu (posts), and HTML or PNG downloads. This addresses a concrete workflow problem: content designed in HTML often needs reformatting when posted to platforms that accept only their own native formats. HTML Anything handles that reformatting step through the export function.

PNG export converts the generated HTML to an image, which is the format required for posting visually designed content to platforms that do not render HTML. The tool's sandboxed preview means the HTML is rendered before export, so the image accurately reflects what the template produces rather than being a blind conversion.

This export capability distinguishes HTML Anything from a plain HTML template library. The templates are designed for export, not just for display in a browser.

## Limitations: Requires a Coding Agent CLI and No Standalone LLM Path

The most significant constraint is the dependency on a pre-installed, pre-authenticated coding agent CLI. HTML Anything has no built-in LLM call path. If you are not already using Claude Code, Cursor, or one of the other seven supported CLIs, you cannot generate content. This is by design: the tool is positioned as a zero-API-key addition to an existing agent workflow, not as a standalone generation tool.

This also means performance and quality depend entirely on the underlying agent's model and session. Different agents produce different results, and a session that has drifted or expired will fail silently or produce degraded output.

The repository has no GitHub releases. The last push was on 2026-09-15. It is not archived. The project is related to nexu-io/open-design, which the README describes as where the same team ships at scale. HTML Anything is described as the focused agent-era HTML editor from that team.

## HTML Anything vs. Building Agent Prompts Manually

An engineer could write a prompt that asks Claude Code or Cursor to generate an HTML document matching a specific visual style. This achieves the same high-level goal but requires writing and maintaining the prompt, describing the layout rules, and specifying color palettes and typography manually each time.

HTML Anything codifies this work into skills. The deck-swiss-international skill, for example, specifies a 16-column grid, a single saturated accent color, and 22 locked layouts. Rather than describing these constraints in a chat message, the skill file provides them to the agent as part of the generation context. The advantage is consistency: each generation follows the same rules without the engineer repeating them.

The trade-off is less flexibility for one-off custom designs. When the skill's locked layouts do not match what you need, you are working around the skill rather than with it. Manual prompting is more flexible for unique layouts; the skill system is more consistent for repeated use of the same template.

## Conclusion

HTML Anything is useful for engineers who already have Claude Code, Cursor, or another supported agent CLI installed and want HTML output instead of plain markdown documents. It is not useful if you have no coding agent CLI set up, since it has no built-in LLM call path. Before using it, verify your agent CLI is on your PATH, as the README states nine CLIs are auto-detected from the PATH.

## FAQ

### What HTML surfaces does HTML Anything support as output?

HTML Anything supports nine deliverable surfaces: magazine articles, keynote decks, resumes, posters, Xiaohongshu (XHS) cards, tweet cards, web prototypes, data reports, and Hyperframes videos. Each surface has one or more skill templates that define its layout and visual rules.

### Does HTML Anything work without an OpenAI API key?

Yes. HTML Anything reuses the authenticated session of whichever coding agent CLI is on your PATH, such as Claude Code or Cursor. No separate OpenAI or Anthropic API key is required by the tool itself, though the underlying agent CLI must already be authenticated.

### What are the nine coding agent CLIs that HTML Anything supports?

The README lists: Claude Code, Cursor Agent, Codex, Gemini CLI, GitHub Copilot CLI, OpenCode, Qwen Coder, Aider, and IBM Bob. These nine are auto-detected from your PATH at startup.

## Sources

- [Issues](https://github.com/nexu-io/html-anything/issues)
- [License: Apache-2.0](https://github.com/nexu-io/html-anything/blob/main/LICENSE)
- [nexu-io/html-anything on GitHub](https://github.com/nexu-io/html-anything)
- [Project website](https://open-design.ai/html-anything/)
- [README](https://github.com/nexu-io/html-anything/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/nexu-io-html-anything