For agents
Give your agent the power to publish to the web.
One skill file. Your agent can publish documents, host live websites, and iterate on both with you through comments. Works with Claude Code, Codex, Cursor, and any agent that can run commands.
Install the skill#
One command, or paste an instruction to your agent. Takes 30 seconds.
Paste this to your agent:
Run this: mkdir -p ~/.claude/skills/html-docs && curl -fsSL https://www.html-docs.com/agents/skill.md -o ~/.claude/skills/html-docs/SKILL.md
Auto-discovered on next session. Then start using the prompts below.
Or use MCP (works with all agents)
{
"mcpServers": {
"html-docs": {
"command": "npx",
"args": ["@html-docs/cli", "--mcp"]
}
}
}6 tools: publish, publish_file, update, read, comment, list_comments.
Use it#
Copy a prompt, swap in your subject. Each one publishes something live.
Implementation notes
A living doc the agent updates as it works — design decisions, tradeoffs, open questions. You follow along and comment in real time.
Implement <SPEC>. As you work, keep a running notes doc on HTML Docs that captures anything I should know — design decisions, deviations from the spec, tradeoffs, and open questions. Create it once, then update it as you go so I can follow along live and comment.
Product requirements doc
Point your agent at a codebase and get a full PRD — personas, user stories, journeys.
Reverse-engineer a product requirements document from this codebase — personas, features as user stories, journeys — and publish it to HTML Docs.
Research report
Deep-dive research on any topic, published as a shareable report with diagrams and citations.
Research [TOPIC] in depth and publish a comprehensive report to HTML Docs — include key findings, analysis, CSS diagrams where helpful, and a summary of implications.
Design review
Get a web page design, then steer it with comments instead of re-prompting from scratch.
Design a landing page for [X] and publish it to HTML Docs. I'll review the design with comments — then read my comments and revise the page from them.
Built with these prompts
Agent–human collaboration#
Share a doc with your agent. Review, comment, and iterate together.
Set your doc’s visibility to public in the share dialog. Anyone (or any agent) with the link can read it. No API key needed.
Pass your API key as x-api-key header. The agent gets full read/write access to your docs — create, edit, comment.
Q3 Product Launch Plan
Ship redesigned dashboard, roll out inline comments...
Target ship date: end of next sprint.
1. Dashboard redesign is the critical path
2. Comments feature unblocks collaboration
3. Timeline is aggressive — needs scope check
API Design Proposal
All endpoints return JSON with a status field.
“Should this be a top-level field or nested under meta?”
Authentication uses bearer tokens...
Rate limiting is set to 1000 req/min.
“Too high for public API — consider 100/min default.”
Introduction
Our company has been working very hard over the past several months to develop what we believe is a truly revolutionary...
We built a faster way to publish docs. Here’s how it works.
Weekly Standup Notes
Replace <doc-id> and <doc-token>with values from your doc’s share URL. With an API key set up, agents can also use the doc ID directly.
Go deeper
The review & iterate loop
How agents create, review with comments, and iterate on docs — the core workflow.
API reference
Full API docs — create, read, edit, comment, version history, authentication.
Full skill file
The raw SKILL.md — view it, copy it, or curl it directly.
CORS proxy for live data
How hosted pages can fetch external APIs through the built-in forward proxy.
Install it once. Your agent publishes forever.
One skill file. Documents and websites. No deploy step, no infra.
Get the skill