Share That Live Dashboard And Your Colleague Sees Their Inbox, Not Yours
Box five on this card is the one marked NEW. “Live Artifacts — live dashboards that pull real data from your connectors and refresh in real time.” The mock-up shows three pills: 247 Gmail · 5 Calendar · 8 Slack. The prompt underneath asks for exactly that: a live status dashboard with the unread count, today’s events and the mentions.
This works. Connector-backed artifacts are real, documented, and genuinely useful. But there is one thing about them that no card mentions and that changes what you would build with them — and you find it the first time you send someone the link.
The page does not call your connectors. It calls the viewer’s.
Source card: “Hidden Claude Features — 6 things you can just say to Claude that most users have never typed” — BANI Academy, 18 October 2026. Memory, Ask First, Interactive Chart, Build a Tool, Live Artifacts, Use a Skill.
View the original post on Facebook
The connector call runs as whoever is looking
From the artifacts documentation, under How connector calls work for viewers:
“Each viewer uses their own connectors: calls go through the viewing account’s connected tools, so two people opening the same dashboard can see different data depending on what their accounts can access. The page never sees anyone’s credentials; claude.ai makes the calls on the page’s behalf.”
“Viewers approve access first: claude.ai asks each viewer for permission before the page’s first connector call. A viewer who declines, or who hasn’t connected a connector the page uses, still sees the page without its live sections.”
“Actions use the viewer’s account too: a page can offer controls that invoke connector tools with side effects… The action goes through the account of whoever selects the control.”
Read that third line twice. If the dashboard has a button on it, the button acts as the person who presses it.
| Who opens your Gmail dashboard | What they see | Why |
|---|---|---|
| You | Your 247 unread, your 5 events, your 8 mentions | Your account, your connectors. Exactly as the card shows |
| A colleague who has connected Gmail | Their unread count, their calendar | The call runs through the viewing account’s connected tools |
| A colleague who has not | The page, with the live sections blank | “Still sees the page without its live sections”. Nothing explains why unless you built a fallback |
| A colleague who declines the permission ask | The page, blank sections, for that page load | “A denial lasts for the rest of that page load. Reloading the page brings the permission ask back” |
| Anyone on a public link, signed out | No live sections at all | “Connector calls don’t run for a viewer who opens the public link without signing in to claude.ai, or from outside your organization” |
Which makes it a template, not a dashboard
That is good news and bad news in one mechanism, and both halves matter.
The good half: you cannot leak your inbox by sharing the page. A connector-backed artifact is not a screenshot of your data; it is a set of instructions for fetching data, executed under whoever opens it. Share it as widely as you like and nobody reads your mail.
The bad half: the thing you built is not the thing you can send. If the point was “look at my numbers”, the link does not carry them. What travels is the shape — the layout, the sections, the logic — which makes a connector-backed page much closer to a template you give your team than a report you circulate.
Once you see it that way it is obviously the better product. “Here is a morning dashboard; open it and it fills with your inbox” is worth more to five colleagues than a picture of yours. The documentation even tells you how to make that land:
“When you plan to share a connector-backed page, ask Claude to include a fallback message in each live section that names the connector it needs. A viewer who’s missing the connection then sees what to connect instead of an empty section.”
One sentence added to the card’s prompt, and the blank-box failure becomes an instruction.
“Real time”, “every hour”, or when it loads?
Box five describes its own refresh behaviour twice, two inches apart, and the two descriptions do not match. The headline says the dashboards “refresh in real time”. The status line under the prompt reads “Claude → Live · refreshing every hour”.
The documented behaviour is a third thing:
“The page fetches data when it loads and can refresh on an interval or when a viewer uses a refresh control on the page. Responses are cached in the viewer’s browser, so a reopened page renders from the cached responses immediately, then updates with fresh results.”
So: on load, by default. An interval is possible if you ask for one, which is where “every hour” could come from. “Real time” is the one description that is not available at all. For an unread-count dashboard that distinction decides whether you leave the tab open or reopen it — and the answer is reopen it, or ask for a refresh button.
Six boxes, three features and a sentence
Step back from box five and count what is actually on the card. The six headings suggest six things to learn. They are not six things.
| The box | What it actually is | What you need before it works |
|---|---|---|
| 01 Memory | A real feature with a settings toggle | Memory switched on. Then one sentence, once |
| 02 Ask First | Not a feature. The card’s own example is a plain English sentence: “Ask me about the topic, format, length, and target audience before writing the script” | Nothing at all |
| 03 Interactive Chart | An artifact | A paid plan |
| 04 Build a Tool | An artifact. The card writes /artifact in the prompt box itself |
A paid plan |
| 05 Live Artifacts | An artifact with the connectors capability declared | A paid plan, plus a connector per data source, per viewer |
| 06 Use a Skill | A real feature. A saved folder of instructions you can trigger again | Somewhere for the skill to live |
And the box that is “just a sentence” is the best one on the card
It would be easy to treat box two as filler. It is the opposite. Ask First costs nothing, needs no plan, no toggle and no connector, and it changes more answers than any of the other five.
The card’s own example is the whole technique: “I want to create an Instagram Reel. Ask me about the topic, format, length, and target audience before writing the script.” Four questions answered up front, instead of a script built on four guesses and then rewritten.
That is worth saying plainly because it inverts the card’s own ordering. The boxes get more impressive as you go down — charts, tools, live dashboards — and more expensive. The cheapest one is the one you should type today, and it works in any AI chat tool, on any plan, with no setup whatsoever.
| Box | What it costs to start | Who it works for |
|---|---|---|
| 02 Ask First | Nothing. One sentence in a prompt | Everyone, including free plans and other chat tools |
| 01 Memory | A settings toggle, then one sentence | Anyone whose account has it |
| 06 Use a Skill | Describing the workflow once | Anyone who repeats a workflow |
| 03 / 04 Artifacts | A paid plan | Pro, Max, Team, Enterprise |
| 05 Live Artifacts | A paid plan, connectors authorised, and a connector per viewer | You alone, unless your colleagues connect theirs too |
One footnote on the artwork: every one of the six prompt boxes shows the model selector reading “Fable 5”, which Anthropic lists as a legacy model. We went through why that matters in the piece on naming a legacy model, and will not repeat it here.
And the card’s fine print deserves credit: “Commands and features change often and may not exist on every plan. Check Claude’s official site for what is available to you. Numbers in the examples are illustrative.” The plan caveat is the right one for a card where three boxes need a subscription, and “numbers are illustrative” covers the 247 unread emails before anyone has to ask.
Four honest caveats
- We did not build the dashboard. The viewer-account behaviour is quoted from the artifacts documentation, not observed. If you are about to depend on it, build one and send it to a colleague before you promise anything.
- “Every hour” may be achievable. The docs say a page “can refresh on an interval”, so an hourly refresh is a thing you could ask for. Our point is that it is not the default, and that “real time” is not on the menu at all.
- Artifact availability is a moving target and varies by plan, organisation policy and admin toggles — there is a separate Enable artifact connectors switch for organisations, independent of artifacts themselves. A colleague seeing blank sections may be looking at an admin setting, not a missing connector.
- This is not a security warning. The mechanism is good: the page never holds your credentials and never carries your data. We are describing a design consequence, not a flaw.
The one-line version
A connector-backed artifact runs its calls as whoever opens it, so the live dashboard you built is a template rather than a report — nobody else can see your numbers in it, and nobody else sees anything in it until they connect their own.
Build it for yourself, add a fallback line naming each connector before you share it, and type box two’s sentence today because it costs nothing.
Source card: “Hidden Claude Features” — BANI Academy, 18 October 2026.
View the original post on Facebook · Related: the twenty-command card, and which of its names the software actually knows
One Comment