Agent skills for unfurl, the single-URL document publisher. Publish Markdown or HTML and get back one unguessable Share link — or update an already-shared doc in place, keeping the same link.
unfurl is the last-mile sharing tool for AI-generated content: paste Markdown or HTML, get back one unguessable Share link, and anyone who opens it can read the rendered page — no login, no app, no download. The link is stable; edit the content and the same link keeps updating in place.
- No login for viewers — open the link and read. No sign-up wall, no pop-ups; the reading experience is the content.
- Markdown or HTML — Markdown is sanitized through a whitelist; a full HTML page renders inside a sandboxed iframe, so neither escapes.
- Unique & unlisted — every link is one of a kind and not indexed by search engines. Only people you hand it to can open it.
- Built to be read — the reading page ships serif headings, light/dark themes, and code highlighting, and it's mobile-friendly.
| Skill | Description |
|---|---|
unfurl-share |
Publish Markdown or HTML to an unfurl instance and get back the Share link, or update an already-shared doc in place. |
unfurl-read |
Read the verbatim content of a shared unfurl Doc from its link — raw Markdown/HTML via the /source endpoint, minimum tokens. |
Install a skill with the open agent skills CLI (npx skills). It auto-detects your coding agent (Claude Code, Codex, Cursor, and 70+ others):
# interactive — pick the skill and the agent(s)
npx skills add jwma/unfurl-skills
# non-interactive examples
npx skills add jwma/unfurl-skills --skill unfurl-share -a claude-code -g -y # Claude Code, global
npx skills add jwma/unfurl-skills --skill unfurl-share -a codex -y # Codex, project-localScope flags:
| Flag | Meaning |
|---|---|
-g, --global |
Install to the user directory (available across all projects) instead of the current project |
-a, --agent <name> |
Target a specific agent (claude-code, codex, cursor, …) |
-s, --skill <name> |
Install a specific skill by name |
-y, --yes |
Skip all confirmation prompts |
--copy |
Copy files instead of symlinking |
List skills without installing, or install all of them:
npx skills add jwma/unfurl-skills --list
npx skills add jwma/unfurl-skills --allunfurl-share reads one environment variable every run. Add it to your shell profile (e.g. ~/.zshrc):
export UNFURL_API_KEY="..." # a Creator's API key from the unfurl dashboard (Dashboard → API keys)The unfurl instance is fixed at https://unfurl.anmuji.com. Set UNFURL_BASE_URL (or pass --base-url) only to target a self-hosted instance.
See the skill's SKILL.md for full usage, flags, and the error-code reference. Reading a shared Doc (unfurl-read) needs no API key — the Share link alone is enough.
Once the skill is installed and UNFURL_API_KEY is set, you don't run any command — you just talk to your agent. It recognizes the intent, runs the skill behind the scenes, and hands back the Share link. It works in any language — English or 中文:
English
| You say … | … and the agent replies |
|---|---|
"Publish notes.md to unfurl as Project notes." |
https://unfurl.anmuji.com/p/abc123 |
"Turn this into a share link: # Sprint 13 retro …" |
https://unfurl.anmuji.com/p/def456 |
"Post this HTML to unfurl: <h1>Demo</h1>…" |
https://unfurl.anmuji.com/p/ghi789 |
"Update abc123 with the latest numbers: # Q3 …" |
https://unfurl.anmuji.com/p/abc123 (link unchanged) |
中文
| 你说…… | agent 回复 |
|---|---|
"把 notes.md 发布到 unfurl,标题《项目周报》。" |
https://unfurl.anmuji.com/p/jkl012 |
"把这个分享成链接:# 第 13 期周报……" |
https://unfurl.anmuji.com/p/mno345 |
"把这段 HTML 发到 unfurl:<h1>演示</h1>…" |
https://unfurl.anmuji.com/p/pqr678 |
"用最新数据更新 abc123:# Q3 ……" |
https://unfurl.anmuji.com/p/abc123(链接不变) |
Or chain it into a bigger task. Describe the whole job and end with the skill — the agent does the work, then publishes the result:
"Analyze this source file, add Mermaid diagrams to illustrate it, write it up as Markdown, then
/unfurl-share."
"分析当前这段源代码,用 Mermaid 画图辅助说明,整理成 md,然后
/unfurl-share。"
The agent builds the writeup and hands back the Share link. (Naming the skill directly with /unfurl-share is just an explicit trigger — plain intent works too.)
It triggers on publish / share / post to unfurl, make a share link, update/edit an already-shared doc (by its share token or link), or any mention of unfurl. Point it at a file, paste the content inline, or just describe what to share; the agent picks the format and runs the script for you.
The underlying flags (--format, --title, --file, …) and the full error-code reference are in SKILL.md — you won't normally need them.
Paste an unfurl Share link into the conversation and ask about it. The agent fetches the Doc's raw content through the dedicated /source endpoint and reads that directly, instead of scraping the rendered page — far fewer tokens, no boilerplate to filter out. No API key is needed to read; the Share link alone is enough.
English
| You say … | … and the agent does |
|---|---|
"Summarize https://unfurl.anmuji.com/p/abc123." |
reads the Doc's Markdown/HTML and summarizes it |
"What does https://unfurl.anmuji.com/p/abc123 actually say?" |
reads and explains it in plain terms |
"Translate https://unfurl.anmuji.com/p/abc123 to English." |
reads the source and translates |
中文
| 你说…… | agent 的反应 |
|---|---|
"帮我总结一下 https://unfurl.anmuji.com/p/abc123。" |
读取原文并总结 |
"https://unfurl.anmuji.com/p/abc123 这篇讲了什么?" |
读取并用大白话讲解 |
"把 https://unfurl.anmuji.com/p/abc123 翻译成英文。" |
读取源文并翻译 |
It triggers whenever you hand the agent an unfurl URL and want the content — summarize, explain, answer questions, translate, or reuse it. See unfurl-read/SKILL.md for details.
Each skill is self-contained under skills/<name>/. Both scripts are Python 3 standard-library only — no dependencies. Run their offline test suites (no network, no credentials):
python3 skills/unfurl-share/scripts/unfurl-share.test.py
python3 skills/unfurl-read/scripts/unfurl-read.test.py