HTML Anything — The Agentic HTML Editor
Turn your local AI coding agent into an HTML typesetter — draft in Markdown, it produces ship-ready HTML, and publish to WeChat, Zhihu, X and more in one click.
Evidence: Sandboxed preview (iframe sandbox) is mentioned, but least privilege principle is not explicitly stated; user confirmation exists (e.g., ⌘+Enter), but details are lacking; data flow transparency is partially addressed (data not uploaded), but not detailed; sensitive data handling is not specified; dependency security lacks vulnerability scanning or lockfile details; external effects (e.g., export to WeChat) are described, but side effects are not covered; rollback is not mentioned; source attribution references multiple open-source projects, but license compatibility is not clarified. Deductions: lack of implementation details, mostly assertions.
Evidence: README is consistent, but code details are missing; dependency availability is not specified; failure messages are not mentioned. Deductions: static review cannot verify runtime behavior, and documentation lacks error handling details.
Evidence: README lists many scenarios and skills, but capability boundaries are not explicit; trigger precision is not detailed; environment fit (e.g., multi-CLI detection) is described. Deductions: capability boundaries and trigger precision lack implementation details.
Evidence: README is well-structured, install notes are detailed, but version history or changelog is missing; known limitations are not explicit; license is Apache-2.0, but copyright info is missing; maintenance responsibility is not clear. Deductions: lack of versioning and maintenance responsibility.
Evidence: Output usability is high (multiple export formats), marginal value is clear (compared to Markdown), cost-benefit is good (zero API key). Deductions: static review cannot verify actual effectiveness, but documentation supports these claims.
Evidence: Some claims are traceable (e.g., references to open-source projects), but cross-source corroboration is lacking; facts and inferences are not clearly separated. Deductions: static review cannot verify claims, and documentation lacks verification methods.
- Static review cannot verify actual runtime behavior; all scores are based on documentation and code structure.
- Dependency security lacks vulnerability scanning or lockfile details; recommend checking dependency versions.
- Rollback mechanism is not mentioned; recommend confirming if undo functionality exists.
- Version history and changelog are missing; recommend checking Git commit history.
What does this agent do, and when should you use it?
HTML Anything is a local-first, agentic HTML editor that reuses the coding-agent CLIs already installed and logged in on your machine (Claude Code, Codex, Gemini CLI, etc.) instead of shipping its own model. It comes with 75 composable skill templates across 9 deliverable surfaces — magazine articles, keynote decks, résumés, posters, Xiaohongshu/Tweet cards, web prototypes, data reports, and Hyperframes video storyboards. The editor streams the agent's output live into a sandboxed iframe preview via Server-Sent Events (SSE), and offers one-click export to WeChat-compatible HTML, high-DPI PNG, or the clipboard. Built on Next.js 16 and React 19, it runs locally with zero API keys and is licensed under Apache-2.0.
After starting the local dev server (pnpm -F @html-anything/next dev), the browser calls /api/agents to scan PATH for installed coding-agent CLIs (including Claude Code, Codex, Cursor Agent, Gemini CLI, etc.). The user picks a template (e.g., Swiss International deck or a Xiaohongshu card), pastes content, and hits ⌘+Enter. The frontend POSTs to /api/convert, which spawns the selected CLI via child_process.spawn and pipes the input to stdin. The CLI's stdout (JSON-lines) is parsed for text deltas and re-emitted as SSE events; the client continuously appends them to the iframe's srcdoc for live preview. Once generation finishes, the user can export: juice inlines CSS for WeChat, modern-screenshot renders the iframe to a 2× PNG and places it on the clipboard (for X/Weibo/Xiaohongshu), or they can download standalone .html and .png files. No API key is required — it reuses the existing CLI sessions.
- Content creators drafting a magazine-style article in Markdown and publishing it to WeChat without manual reformatting.
- Product managers generating structured internal docs (PM spec, OKRs) in elegant HTML for team review.
- Marketers creating Xiaohongshu cards or Weibo visuals and exporting to 2× PNG for direct posting.
- Developers needing a quick web prototype to showcase an idea, with interactive preview in the sandboxed iframe.
- Trainers and educators preparing slide decks with speaker notes and PDF export for classroom use.
- Video producers creating Hyperframes storyboard frames (glitch titles, brand outros) for Remotion video rendering.
What are this agent's strengths and limitations?
- Zero API keys: reuses already-authenticated CLI sessions (Claude, Codex, Gemini, etc.) — no extra cost.
- Supports 9 major coding-agent CLIs with automatic PATH detection, no manual configuration.
- 75 skill templates enforce hard design constraints (CJK-first fonts, 8px grid, contrast ≥4.5) for consistent, high-quality output.
- One-click export to WeChat, Zhihu, X and more, saving significant formatting time.
- Requires an active subscription to at least one coding-agent CLI (e.g., Claude Pro, GitHub Copilot) to generate content.
- Local environment dependency; adapter maintenance may be needed as CLI versions evolve.
- SSE streaming can be affected by CLI output format; abort-and-retry isn't fully polished.
- Advanced features like multi-template comparison and history are still on the roadmap.
How do you use this agent?
Clone and enter the repo: git clone https://github.com/nexu-io/html-anything && cd html-anything. Install dependencies: pnpm install. Start the dev server: pnpm -F @html-anything/next dev. Open http://localhost:3000 in a browser — the top bar auto-detects installed coding-agent CLIs (must be pre-logged-in, e.g., claude login). Pick a skill template, paste content, and press ⌘+Enter to generate.