# HTML View

> Create or revise live .univer.html pages backed by existing Workspace Sheet data using HTML bindings and JavaScript. Validate sources and publish pages with HTML View operations; revise existing pages through the Blob workflow.

- Skill: `dream-num/html-view` (Agent Skill, multi-file: 16 files)
- Install (CLI): `npx get-skill add dream-num/html-view`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dream-num/html-view/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Web & Frontend
- Author: dream-num (https://skillmd.com/u/dream-num)
- Updated: 2026-11-15
- Page: https://skillmd.com/skills/dream-num/html-view

---


# Workspace HTML Views

With the CLI, `html-view create` and replacement publish immediately. Through MCP tools,
creation and replacement save a draft the user publishes in Workspace. Follow the server instructions.

## Start with a style and live data

1. Read [design.md](references/design.md). Read the source Sheet, then, unless the user
   already named a style, ask which style they want as design.md describes, in the same
   message as your plan for the page. End your turn and wait for the answer before you write,
   validate or publish a page. If they leave the choice to you (你来定, 随便, 默认就行), use
   Night Gauge (`night-gauge`, 仪表夜航). Never silently choose another style.
2. Read and adapt the complete `templates/<id>.univer.html` for that style; never start from
   a blank page. Keep its fonts, palette, density and signature element when adapting to new data.
   For a style the user describes, start from the template whose structure best fits the data and
   replace its fonts, palette and signature element with the user's description.
3. Page controls write to Sheet cells or append rows; the page displays the recalculated
   source values through live bindings, and label host-provided visitor trials with `getWriteMode`
   as the templates do, without page-side simulations. Never re-implement a Sheet formula in page code,
   embed copies of Sheet data, or use fallback numbers. Keep loading, empty and error states honest.
4. Write the page in the user's language as [design.md](references/design.md) describes;
   templates are English starting points.

## Bind the page

Use HTML, CSS and JavaScript to present or edit existing Sheet data in a Workspace page.
Declare each live component's source with `data-univer-cell-subscribe` or
`data-univer-range-subscribe`, then register `subscribeCellById()` / `subscribeRangeById()` callbacks.
Use fine-grained declarations on the consuming elements so bindings are traceable in HTML;
plain text and native controls use `data-univer-cell-text` / `data-univer-cell-model`.
For binding syntax, JavaScript data access, controls and frontend libraries, load the [authoring.md](references/authoring.md) reference of the `html-view` Skill and its linked references as needed. The `core` Skill explains how to load them. If the needed sections are already in context, use them directly.

Follow the `core` Skill for source and asset handling. `window.univerBinding` is provided at the published Workspace URL.

## Find the sources

Use the `core` Skill for authentication and discovery, and the `sheet` Skill to inspect worksheets, values and formulas. Bind to actual Unit
and worksheet IDs; Resource IDs and worksheet names are different identities.

Sources always resolve on trunk. If the required Unit, worksheet or changes exist only in a
Worktree, prepare the template but defer validation and publication until those changes are
merged through the existing review workflow. Do not automatically merge a Worktree or substitute
older trunk data. Creating an HTML View does not authorize changing its source cells or formulas.

## Host navigation

Workspace renders pages in an iframe with `sandbox="allow-scripts allow-forms"` and blocks native form navigation. Handle form submission with `preventDefault()` and the binding API. Popups and top-level navigation are unavailable; a normal link navigates inside the iframe. Return separate Workspace page links to the user instead of adding cross-page navigation that loads Workspace inside the view.

## Validate and create

Run `html-view validate` with the complete HTML source. Then run `html-view create` with the same source, destination Space, name `Dashboard`, stable idempotency key, and optional parent Node. Follow the `core` Skill for source handling.

`validate` parses declarative HTML bindings, reads accessible trunk Sheet data, and checks
worksheet existence and cell/range bounds. It returns `{ valid, unitIds, bindingCount, bindings, truncated? }`, with at most 60 binding samples of cell values and range summaries in document order (`truncated: true` when more exist).
Compare `bindings` with the page plan for empty cells where data is expected, header text where numbers are expected, formula cells behind controls that write, and numbers carrying a date format.
It does not execute JavaScript or discover references created by scripts. A JavaScript-only page may report
zero static bindings; runtime requests still require access to each source Unit.

Blob write keys must be 16–200 ASCII letters, digits, underscores or hyphens; a UUID is a suitable key. This applies to both creation and replacement.

`create` performs the same validation and uploads the validated HTML as a new Blob. It adds
`.univer.html` to the name if missing and returns Blob identities plus `workspaceUrl`, `shareUrl`, `repositoryVisibility` and `privateSourceUnitIds`. The Space
is explicit; the parent defaults to its root. Reuse an idempotency key only when retrying identical
content, name and destination. Creation publishes immediately without a Unit Worktree and does
not modify source cells.

Review the template's structure, script syntax, referenced IDs, coordinates, loading/error states
and intended write handlers. Template validation does not prove runtime rendering, interaction or
synchronization; report which checks ran. Unit screenshot commands do not render HTML Views.

## Share the page

Return `shareUrl` (the immersive page link) to the user and say who can open it.
When `repositoryVisibility` is `public` and `privateSourceUnitIds` is empty, anyone
with the link can view the page and its declared sources. Otherwise, only signed-in
people with access can open the page with all its data. A public page can open
without its private source data. Signed-out visitors and signed-in non-members are
read-only; page writes are rejected for them.

If the user wants anyone with the link to open a private repository's page, ask
before making the repository public: this immediately publishes everything in the
repository. After approval, use CLI `repo edit <repo> --visibility public`, or MCP
`repo_edit {repo, visibility: "public"}` in the `admin` toolset.
Keep source Sheets in the same public repository as the page, or list the
`privateSourceUnitIds` to the user. These fields describe declared sources at
creation time; script-created references and later permission changes still need
checking, and Workspace authorizes every visitor's requests.

Media referenced with `data-workspace-src` appears only when the file itself is in a public
repository or has link sharing on, for every visitor including signed-in members. Tell the user
which referenced files are private instead of changing their sharing without asking.

## Revise an existing page

Read the existing page's HTML and strong ETag, edit the original source, validate the
complete HTML, then replace it with that ETag and a new stable write key. Use the
`core` Skill's operation bindings for the connected product. With the CLI, read the
`blob` Skill, download the Resource, run `html-view validate`, then use `blob replace`.
Replacement preserves the existing page identity and URL. Reconcile ETag conflicts;
create another Blob only when a separate page is wanted.

Page access does not grant access to the source Sheets. Author validation uses the current authenticated
identity; each viewer's source access is checked separately by Workspace at runtime.

