"Uploads the file below from this task to claude.ai": Claude's Publish Artifact Prompt Explained
What Claude's Publish artifact prompt means: "Uploads the file below from this task to claude.ai", who can see the page, the three variants, and how to undo it.
Short answer
It is Claude Code asking before it publishes an artifact: it uploads the HTML or Markdown file shown below the prompt to a page on claude.ai. A new artifact is private, so only you can open it until you use Share. If the same file was already published in this task, approving updates that page instead and keeps who can see it. Approve it if the file holds nothing you would not store with Anthropic.
Want the page on your own domain instead?
Connect Claude Code to VibeDeploy and say "deploy this page": the same file goes live on a normal website anyone can open, with your own domain one step away. 14 days free, no card.
claude mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcpYou asked Claude to build something visual, and instead of the result a permission prompt appeared: Publish artifact, followed by a sentence like "Uploads the file below from this task to claude.ai." It is fair to stop and ask what that means before you press approve. Here is each version of the prompt in plain language, what happens when you approve it, who can see the page, and how to undo it.
What the Publish artifact prompt is
An artifact in Claude Code is a live web page that Claude publishes from your session to claude.ai. Claude writes the page as an HTML or Markdown file, usually in a temporary folder outside your project, and then publishes it. The page gets an address that starts with claude.ai/, opens in your browser, and updates in place when Claude publishes again.
Most of the time this happens without a question: according to Claude Code's documentation, a new artifact that Claude publishes in reply to a prompt you typed goes through without a permission prompt. You see the Publish artifact prompt when Claude Code wants your approval first. The sentence under the title tells you which kind of publish it is. There are three.
Prompt 1: "Uploads the file below from this task to claude.ai"
Uploads the file below from this task to claude.ai. New artifacts stay private until you share them; if this file was already published in this task, that artifact is updated instead and keeps its current sharing settings.
This is the ordinary publish. Approving it does two things, depending on the file:
- A file published for the first time becomes a new artifact that only you can see. Nobody else can open the link until you share it.
- A file that was already published in this task updates the page it created before, at the same address. Whoever could open it before can open it now, and they see the new version.
"The file below" is the file named under the prompt. That file, the page itself, is what gets uploaded; the prompt does not upload your project folder.
Prompt 2: "...and replaces what an existing artifact shows now"
Uploads the file below from this task to claude.ai and replaces what an existing artifact shows now. That artifact keeps its current sharing settings, so anyone who can already open it will see the new version.
This one updates an artifact that already exists, for example a page you asked Claude to update by giving it the artifact's link, or one you shared publicly: Claude Code asks once per conversation before it changes a publicly shared artifact. Pay attention to the second sentence. Sharing is not reset, so if the page is shared with your team or on a public link, the new content is visible to all of them as soon as it is published.
Each publish is kept as a version. From the Share control in the page header you choose which version viewers see, so a wrong update can be rolled back there.
Prompt 3: "Creates a new artifact on claude.ai from the artifact type below"
Creates a new artifact on claude.ai from the artifact type below. Its page was written by that type's publisher and no file is uploaded from this task; the new artifact stays private until you share it.
Here Claude starts from a ready-made artifact type instead of a page it wrote. Claude Code's documentation names three templates of this kind: Claude Slides for a presentation, Claude Design for a visual design and Claude Docs for a document. The page code comes from the type's publisher, Claude fills in the content, and nothing from your project folder is uploaded. Like any new artifact, it is private until you share it. The templates are in beta and on by default on Pro, Max and Team; on Enterprise, an Owner turns each one on.
When Claude Code asks before publishing
If you rarely see the prompt, that is expected. The documentation lists the cases where Claude Code asks first, or in auto mode lets its classifier review the publish:
- The session is in plan mode: it asks before the first publish of each artifact.
- The page declares runtime capabilities, such as live calls to your claude.ai connectors or a file download button.
- You shared the artifact publicly: it asks once per conversation before changing it.
- A scheduled task started the turn, not a prompt you typed.
- Claude can't read the file without your approval, for example a file outside your working directories.
- In Manual or acceptEdits mode, publishing includes a small edit to the file that needs your approval. The prompt then ends with a sentence such as "Also makes 1 edit to that file on disk."
Is it safe to approve?
Approving does not make anything public. A new artifact is visible only to you, and Anthropic stores artifact content on its own infrastructure. The real decision comes later, when you press Share.
Before you share, look at what is on the page. Claude builds artifacts from anything your session can reach, including your codebase and data it pulled through your connected tools. A dashboard that is fine for you can contain customer names, internal numbers or a stray API key that should not travel further. Read the page as its future viewers will, then share.
If the prompt mentions runtime capabilities, the page will call your claude.ai connectors each time someone opens it, through the account of whoever is viewing. That is useful for a live team dashboard; just know the page is not a static snapshot.
Who can see the artifact after you share it
A new artifact is visible only to you. To share it, open it and use Share in the page header. What you can choose depends on your plan:
- Pro and Max: a public link that anyone on the internet can open, without signing in to claude.ai.
- Team and Enterprise: specific people in your organization or everyone in it, who sign in to claude.ai to view it. A public link is possible only after an Owner turns on external sharing for the organization; it is off by default.
On a public link, viewers who are not signed in, or who are outside your organization, see the label "Content is user-generated and unverified." instead of your name. If an artifact made in a claude.ai chat won't share, or the Publish button seems to be missing, see why a Claude artifact won't publish or share.
How to find, undo or turn off artifacts
- Find it again: run
/in Claude Code to list your artifacts and the ones shared with you. Pressartifacts oto open one,cto copy its link.Ctrl+]reopens the session's most recent artifact, and every artifact you created is in your gallery at claude.ai/code/artifacts. - Delete it: in
/, select it and pressartifacts d, then confirm (outside a cloud session). This permanently deletes an artifact you own, and the link stops working for everyone. - Roll back an update: pick an earlier version under Share.
- Turn artifacts off: switch off the Artifacts row in
/, setconfig "enableArtifact": falsein your settings, setCLAUDE_CODE_DISABLE_ARTIFACT=1, or addArtifacttopermissions.deny. Claude then writes a local file or tells you it cannot publish.
Want the page on your own domain instead?
Connect Claude Code to VibeDeploy and say "deploy this page": the same file goes live on a normal website anyone can open, with your own domain one step away. 14 days free, no card.
claude mcp add --transport http --scope user vibedeploy https://mcp.vibedeploy.be/mcpWhen an artifact is the wrong home for the page
An artifact is built for showing work: a pull request walkthrough, a dashboard, a set of options to compare. Claude Code's documentation is explicit about what it is not. It is one self-contained page with no backend, so it cannot serve multiple routes, relative links do not resolve, the rendered page must stay under 16 MiB, and for a hosted tool with a backend the documentation tells you to deploy it on your own infrastructure. It also lives on claude.ai: there is no custom domain, and on Team and Enterprise plans viewers sign in unless an Owner allows public links.
When the page should be a real website, publish the same file on a web host:
- Connect VibeDeploy to Claude Code once:
claude mcp add --transport http --scope user vibedeploy https://, then runmcp.vibedeploy.be/ mcp /to sign in. The 14-day trial needs no card.mcp - In the session that made the page, say "deploy this page as my-page". Claude sends the file and replies with a live HTTPS address on vibedeploy.be.
- Add your own domain in the dashboard and set the DNS records it lists; the certificate is issued automatically.
Anyone can open that address without a Claude account, and every deploy is kept as a version you can restore. One caveat: a page that relies on artifact-only features, such as live connector calls through claude.ai or the download capability, needs claude.ai to work. Ask Claude for a static version before you deploy it elsewhere. If you would rather move the file yourself, save it as index.html and drop it into a new site in the dashboard. The terminal workflow from an empty folder to a live site is in the Claude Code guide, and the connector setup for Claude chat is in deploy a website from the Claude connector.
Sources
- Claude Code Docs, Share session output as artifacts: what an artifact is, when Claude Code asks before publishing, the "Also makes 1 edit to that file on disk." sentence, updating and versions,
/and its keys, sharing per plan, the unverified-content label, templates, page constraints and the 16 MiB limit, turning artifacts off, storage on Anthropic-operated infrastructure. Checked 26 October 2026.artifacts - The three prompt texts are quoted from Claude's Publish artifact permission prompt as users report and search for it. Checked 26 October 2026.
Ship your AI-built site in minutes
VibeDeploy hosts your AI-built websites in the EU with custom domains, automatic SSL, and a 14-day free trial that gets you online today.
Deploy guides for this topic
Questions people ask
What does "Uploads the file below from this task to claude.ai" mean?
Claude Code is about to publish an artifact: it sends the file named below the prompt, an HTML or Markdown page it wrote, to claude.ai, where it becomes a live page at an address that starts with claude.ai/code/artifact/. The rest of the sentence explains who can see it: a new artifact stays private until you share it, and a file that was already published in the same task updates its existing page.
Can other people see my artifact after I approve?
Not yet. Claude Code's documentation says a new artifact is visible only to you. Other people see it only after you open the page and use the Share control in its header: on Pro and Max you can share a public link that opens without signing in, and on Team and Enterprise you can share with people in your organization, or publicly once an Owner has turned on external sharing.
Why does Claude usually publish without asking?
A new artifact that Claude publishes in reply to a prompt you typed goes through without a permission prompt. Claude Code asks first in specific cases: in plan mode, when the page declares live connector calls or file downloads, before changing an artifact you shared publicly, when a scheduled task started the turn, when Claude needs your approval to read the file, and when publishing includes an edit it needs your approval for.
How do I delete an artifact I approved by mistake?
Run /artifacts in Claude Code, select the artifact and press d, then confirm. That permanently deletes an artifact you own, and its link stops working for everyone, including you. If you only want fewer people to see it, change the audience under Share on the page instead.
Can I stop Claude Code from publishing artifacts?
Yes. Turn the Artifacts row off in /config, set "enableArtifact": false in your settings, set the environment variable CLAUDE_CODE_DISABLE_ARTIFACT=1, or add Artifact to permissions.deny. Claude then writes a local file or tells you it cannot publish.
Can an artifact have its own domain?
No. An artifact is a page on claude.ai: one self-contained page without a backend, up to 16 MiB. To serve the same HTML at your own address, publish the file on a web host. With the VibeDeploy MCP server connected, Claude Code can do that in the same session.
Related reading
Claude Artifact Won't Publish or Share? Causes and Fixes
Why a Claude artifact won't publish or share: Claude accounts, admin settings, the 16 MiB limit, the renamed button. How to unpublish, and the fix.
/ home/ workdir/ artifacts/ index.html: How to Get Your File Out
Claude saved your site to /
From Claude Code in your terminal to a live site
Yes, Claude Code can deploy a website: one claude mcp add command, a sign-in with /
Claude Artifact Custom Domain: Download, Host, and Launch
Download or copy your Claude artifact code, host it outside Claude, and connect a custom domain with EU hosting and automatic Let's Encrypt SSL included.