Answer me with HTML
Turn complex questions into readable, self-contained HTML pages that work offline.
- Source repo
- QingYunA/answer-me-with-html
- Stars
- ★ 6.6k
- Last updated
- 2d ago
- License
- MIT
- Primary language
- JavaScript
- FA score
- 69/100 · Some gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A developer learning the TCP handshake can get a page with a sequence diagram, state diagram, and flag table.
- Not a fit if
- Users who only need short, plain-text answers
- Website
- answer-me-with-html.com ↗ set on GitHub
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-20latest release v0.6.2 (2026-11-20)+38.6 ★/day over 7 days
- Source review
- 69/100 · Some gaps
What does this agent do, and when should you use it?
Answer me with HTML is an agent skill for Claude Code, Codex, Cursor, OpenCode, Pi, and other assistants, with a terminal CLI named am. The assistant writes a Markdown draft, and the CLI renders it as a single-page HTML document with text, tables, code, and diagrams. Components include flow, sequence, tree, and timeline, with blueprint, shadcn, and paper themes. Each output is a standalone HTML file that works offline and embeds its Markdown source. It requires Node.js 20 or newer; the README does not specify an API key or model-service pricing.
After installation, ask an assistant to explain a concept, map a codebase, or compare options; the assistant decides whether a page is useful, drafts Markdown, and invokes the am CLI. The CLI parses frontmatter, components, and code blocks, computes panel and diagram layouts, applies a theme, and writes one HTML file, by default under ~/.answer-me-with-html/pages/. Code blocks can quote actual project files with src= and lines=; the CLI refuses files outside the folder and files that hold keys. For video requests, am video turns a draft with narration beats into a player page, and --mp4 can produce a video file.
- A developer learning the TCP handshake can get a page with a sequence diagram, state diagram, and flag table.
- An engineer onboarding to an unfamiliar repository can map its folders and module call relationships.
- A team choosing between Redis and Memcached can review a comparison table and verdict.
- A developer planning a cache change can inspect actual code alongside open decisions presented as options.
- A technical educator can request an explainer video with step-by-step diagrams and narration.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npx skills add QingYunA/answer-me-with-html
- Config files
~/.claude/skills/answer-me-with-html~/.answer-me-with- Runtime
- Node.js 20
How do you install or deploy this agent?
Requires Node.js 20 or newer. Installing the skill does not require a separate npm install. The README recommends asking a supported assistant to follow INSTALL.md; it also documents the general skills installer and Claude Code plugin commands. The general installer asks which assistants to target. For the Claude Code plugin, run these inside Claude Code:
npx skills add QingYunA/answer-me-with-html/plugin marketplace add QingYunA/answer-me-with-html
/plugin install answer-me-with-html@answer-me-with-htmlYou can also copy skills/answer-me-with-html into an agent's skill folder. The README gives this Claude Code example:
git clone --depth 1 https://github.com/QingYunA/answer-me-with-html.git /tmp/answer-me-with-html
cp -R /tmp/answer-me-with-html/skills/answer-me-with-html ~/.claude/skills/answer-me-with-htmlHow do you use this agent?
After installation, ask the assistant a question that benefits from a visual explanation, for example:
Explain the TCP three-way handshakeYou can also ask explicitly for HTML, or use the CLI to inspect, change, or reset settings:
am config
am config set open off
am config resetPages are saved under ~/.answer-me-with-html/pages/ by default. For a video, explicitly ask the assistant, for example:
make a 3b1b-style video on the TCP handshakeThe README says am video renders the video draft and --mp4 adds a video file output.
What are this agent's strengths and limitations?
- The assistant writes Markdown rather than hand-authoring markup; the README benchmark reports about one-eighth as many output tokens for the same questions.
- The CLI computes panel and diagram layout and provides components such as
flow,sequence,tree, andtimeline. - Each page is a standalone HTML file with no CDN links or web fonts, so it opens offline and is easy to share.
- Pages embed their Markdown source and support comments and selectable answers on the page.
- The README explicitly lists support for Claude Code, Codex, Cursor, OpenCode, and Pi.
- Requires Node.js 20 or newer and an assistant able to run shell commands and access files.
- Requires installing a skill or the Claude Code plugin; it is not a standalone hosted Q&A service.
- The README does not specify model/API charges or a model provider, so total usage cost depends on the assistant used.
- The ASD-STE100-based writing check covers documented machine-checkable rules and warns by default.
How does this agent compare with similar options?
The README compares the project with Archify and GenUI plugins such as dsh-genui. Archify creates an interactive diagram from a JSON spec, while GenUI plugins show components inside one chat app. This skill renders a complete page with text, tables, code, and diagrams from a Markdown draft, for use with an assistant that can run a shell command.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Answer me with HTML This agent | 69 · Some gaps | Agent plugin / skillFree + model costs | ★ 6.6k | 2d ago | JavaScript | Codex · Claude Code |
| Markdown Viewer Agent Skills | Insufficient evidence | Agent plugin / skillFree | ★ 3.4k | 1mo ago | JavaScript | Codex · Claude Code |
| AgentCairn | 89 · Good | Agent plugin / skillFree | ★ 78 | 2d ago | Python | Codex · Claude Code |
| Excalidraw MCP Server & CLI | 73 · Some gaps | CLIFree | ★ 2.7k | 12d ago | TypeScript | ChatGPT · Codex · Claude Code · Claude.ai |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes offline single-file pages and says source quoting refuses files outside the project folder and files that hold keys. Page cleanup asks first, providing some least-privilege, sensitive-data, and deletion-confirmation safeguards. Installation and updates contact GitHub; the CLI can open a browser and create files, while video features can invoke TTS or encoding tools. These effects and parts of the data flow are described, but the packet does not establish implementation details, per-action consent, or a complete recovery path, so those criteria do not receive full marks. Dependencies are listed and bundled, but no dependency-risk audit evidence is supplied, so dependency security scores 1. Repository, package, and MIT attribution are traceable, while maintainer identity and responsibility are not fully clear.
README, SKILL, package metadata, and the component registry broadly agree that the model writes a Markdown draft and the CLI renders it; the README gives the Node.js 20+ requirement and bundled CLI/dependencies. CI configuration covers several Node versions and platforms, build, coverage, install smoke checks, and browser/video workflows; test files show concrete edge-case and error assertions. This is still a static review and cannot establish that those workflows passed. Some features also depend on external Node, browser, or optional video tools being available.
The SKILL and README cover explanations, architecture, comparisons, diagnosis, flows, hierarchies, history, video, and settings. They describe when to skip pages and document components, languages, and themes. Boundaries such as extra tools for video and English-button fallback for other languages are stated. The trigger guidance both encourages broad proactive use and gives narrower page criteria, and the packet does not establish full compatibility boundaries for every advertised host.
Repository organization is clear. The README includes installation methods, settings, examples, components, reference material, and development notes; the MIT license is explicit. The package, plugin manifest, and skill use stable names and versions, but a README release example does not match the supplied 0.5.11 manifest, and the packet contains limited release-history detail. Documentation says older versions are unsupported for security fixes, updates are manual, and gives a vulnerability-report path, but maintainer responsibility is not fully explicit. FAQ and limitation coverage is useful but incomplete.
The product creates a readable offline HTML page with diagrams, themes, quoted code, comments, and interactive decisions; examples make the intended output concrete. The README reports benchmark results and points to corpus material, suggesting lower model output tokens and wait time. The sample is small, results depend on the model and questions, and extra page generation may not help every task, so marginal value and cost-benefit do not receive full marks.
Important claims can be mapped to the README, SKILL, package metadata, component registry, tests, and CI configuration. Some benchmark claims point to corpus data and downloadable pages, and test source includes inspectable assertions. There is some corroboration across files, such as the Node requirement and package engine metadata, but performance, broad host compatibility, and security claims are partly documentary; static review cannot confirm runtime results. The documentation separates reported measurements from feature descriptions to some extent, but does not consistently establish measurement scope or inference limits, so these criteria receive moderate scores.
- The performance figures come from a small benchmark described in the README and should not be generalized to other models, questions, or host environments.
- The plugin/skill can write page files, contact GitHub for update checks, and open a browser when configured; confirm the target agent's install location and local data handling before installation.
- The packet says code quoting skips key files, but does not provide enough implementation source to independently review the full file-filtering behavior.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
<!-- release 0.6.2 2026-11-20 --> Export menu: copy as image, download PNG, save as PDF The toolbar's Copy as image button is now an Export menu (#433, #442): - Copy as image puts the whole page on the clipboard as a PNG. - Download PNG saves the same image as a file named from the page title. The code comes from #434 by @Linuxdazhao. - Save as PDF opens the browser's print dialog with the page's print styles. …View on GitHub ↗ -
Release notes excerpt
<!-- release 0.6.1 2026-11-16 --> New diagrams: state machines, UML classes and API contracts - state draws a state machine: one From -> To: event per line, [*] for the initial and final states, and a loop for Open -> Open: retransmit. A TCP handshake or an order lifecycle reads as states and events instead of a generic flowchart (#388, #390). - class draws UML class diagrams: attributes and methods, interfaces, abstract classes and enums, with inheritance, realization, composition, aggregation, association and dependency (#386, #387). …View on GitHub ↗ -
Release notes excerpt
<!-- release 0.5.11 2026-11-05 --> Send a page to a chat app: am export Discord, Telegram, Slack and WeChat show an .html file as source. am export turns a rendered page into files they show (#350, #354): ```bash am export page.html --pdf page.pdf # A4, the page's own print rules am export page.html --png out/ # out/page-1.png, out/page-2.png ... ``` - --png renders at 1200 CSS px wide and 2x, hides the toolbar, the reply and remark controls and the new-version notice, and cuts the page into slices no taller than 1800 CSS px. A cut falls in a gap between panels. …View on GitHub ↗
FAQ
Do I need a paid API key?
Do generated pages need an internet connection?
Can it quote real code from my project?
src= and lines= to quote file content. Files outside the folder and files that hold keys are refused.Does every question create a page?
Can it make videos?
am video renders a draft with narration beats as a player page, and --mp4 can add a video file.Is Answer me with HTML open source?
Is Answer me with HTML safe to use?
Is Answer me with HTML still maintained?
Who makes Answer me with HTML?
What is Answer me with HTML's official website?
Thanks — we use this to improve the page.