collaborate-dev — Claude Code skill: features and setup
collaborate-dev is a Claude Code skill that gives agents and people a shared workspace for notes, HTML, and diagrams. It activates for visual output or live co-editing.
MIT 1 installs updated 2026-07-28
npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a claude-codeWhat collaborate-dev does
collaborate-dev gives an agent and its human collaborators a shared desktop on collaborate.dev. The agent can create Markdown notes, HTML pages with previews, and canvas elements. People can edit them, comment, add context, and rearrange the layout; everyone sees changes live.
To start, the agent calls the API to reserve a desktop and receive a key, then follows the instructions returned in the response. No local server is needed. A desktop without an account lasts 48 hours from creation, and activity does not extend that window. HTTP clients other than curl need a real User-Agent header; default agents in the Python-urllib style receive a 403 response.
What it helps with
- put notes and HTML artifacts in one shared workspace
- preview HTML without running a local server
- lay out alternatives side by side on a canvas
- coordinate parallel agents through a shared status board
Example prompts for the agent
- Build an HTML dashboard in the shared workspace and show its preview.
- Draw a diagram on the canvas and leave room for comments.
- Create a status board for parallel agents working on this task.
How to install collaborate-dev
1Skills CLI (Vercel)
npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a claude-code2Manually: put the files in place
~/.claude/skills/collaborate-dev/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/collaborate-dev/
1Skills CLI (Vercel)
npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a codex2Manually: put the files in place
.agents/skills/collaborate-dev/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a cursor2Manually: put the files in place
.cursor/skills/collaborate-dev/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install collaborativecomputing-ai/collaborate-dev-plugin skills/collaborate-dev2Manually: put the files in place
.github/skills/collaborate-dev/SKILL.md1Skills CLI (Vercel)
npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a gemini-cli2Platform CLI
gemini skills install https://github.com/collaborativecomputing-ai/collaborate-dev-plugin --path skills/collaborate-devManually: put the files in place
.devin/skills/collaborate-dev/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about collaborate-dev
How do I install collaborate-dev in Claude Code?
Run npx skills add collaborativecomputing-ai/collaborate-dev-plugin --skill collaborate-dev -a claude-code. The skill file is at ~/.claude/skills/collaborate-dev/SKILL.md.
What does collaborate-dev do?
It gives agents and people a shared desktop for Markdown notes, previewable HTML, and canvas work. Collaborators can edit the same materials live.
What does collaborate-dev need to run?
It reserves a desktop and key through the collaborate.dev API. The hosted workspace needs no account; HTTP clients other than curl must send a real User-Agent header.
Source text
SKILL.md · MIT
---
name: collaborate-dev
description: Give this agent a shared desktop for agent work - a live collaborate.dev workspace where it builds markdown notes, HTML artifacts, and diagrams while humans edit, comment, sketch, and paste in context alongside it. Use when the user wants output outside the terminal or laid out visually - "show me", "side by side", "make a dashboard", "sketch / diagram / whiteboard this", "preview this HTML" with no local server - or wants a desktop their team and agents build on together, organize artifacts spatially, co-edit on a canvas, or run a status board that parallel agents coordinate through on a long task.
---
# collaborate.dev - a shared desktop for agent work
collaborate.dev is a shared desktop you write to over HTTP: markdown notes, live HTML artifacts that
render in a sandboxed iframe, and canvas primitives. Mutations flow in every direction, not just
from you: the humans on the desktop edit your notes, comment on them through the commenting system,
paste in URLs as context, sketch, add notes, rearrange the layout, and edit code while you work.
Every collaborator - human, another agent, or a swarm of agents - sees the same state, synchronized
live.
Reach for it when the work is easier to see than to scroll: comparing outputs side by side, laying
out a plan or a diagram, building and previewing HTML with no local dev server, or working through
something on a canvas with a human.
## Step 1 - get your own desktop
You need a desktop and a key. One call mints both, with no account and no auth:
```bash
curl -sS -X POST https://api.collaborate.dev/v1/remote/desktops/reserve \
-H "content-type: application/json" \
-H "user-agent: claude-code-collaborate-dev/1.0" \
--data '{"intent":"copy_prompt"}'
```
The JSON response contains:
- `prompt` - your complete working instructions, with your key already filled in. This is the field
that matters.
- `urls.view` - a read-only link for anyone who should follow the work live without editing.
- `urls.edit` - the full-access link. Your bearer token is embedded in it, so treat it like the
credential it is.
## Step 2 - follow the returned prompt
The `prompt` field is a full guide to reading and writing the desktop: how to store your key, create
and patch notes and HTML, use the REPL for batches and canvas primitives, and the rules for
collaborating with the humans and agents editing the same elements - the comment threads, proposals,
and version guards it describes exist because your collaborators are active, not spectators. Read it
and follow it. It is always current, so this skill does not repeat it.
Do two things right away:
1. Hand the human the right link, and ask them to open it so they can build alongside you from the
first artifact:
- Give the person you are working with the **edit** link (`urls.edit`) when they may want to
edit alongside you or keep the desktop - claiming it (Share → Make private) lives there, and
it holds the same token you already use, so nothing new is exposed.
- Give the **view** link (`urls.view`) to anyone who should only follow along.
- Tag whichever link you hand over so collaborate.dev can attribute it to this plugin: insert
the query string before the `#`, turning
`https://collaborate.dev/remote/view#token=…` into
`https://collaborate.dev/remote/view?utm_source=claude-code&utm_medium=plugin#token=…`.
The tag names the channel only; it carries nothing about you, your machine, your code, or
your session.
2. Set up your key as the returned prompt describes (a header file kept outside the working
directory), then create your first artifact early and keep patching it, rather than building
everything offline and revealing a finished dump.
## Notes
- Expect the desktop to change between your writes: human edits, comments, pasted context, and
moved elements are part of the work, not interference. Re-read before you overwrite, and treat
a version conflict as a collaborator's change to carry forward, not an error to force past.
- The workspace is free and needs no sign-in. A reserved desktop lasts **48 hours from creation** -
a fixed window that activity does not extend. To keep it longer, the human opens the edit link,
signs in, and claims it (Share → Make private). Claiming revokes the public links and your token,
so finish or checkpoint work first; the owner can then issue you a fresh agent key from
Sharing → Agent instructions.
- The desktop is hosted: no local server, build step, or install - it is your run and preview
surface, and you reach it over plain HTTP from wherever you already run. Your first request may
trigger a permission prompt for the API domain; that is expected.
- If you use an HTTP client other than curl, send a real User-Agent header. Default library agents
(Python-urllib style) are rejected with 403.
© collaborativecomputing-ai · MIT · collaborativecomputing-ai/collaborate-dev-plugin