Sharepoint HTML

Publishing HTML to SharePoint (OneUp HTML viewer)

SharePoint already hosts HTML: save an .html file into OneDrive or SharePoint and open it, and it renders in the OneUp HTML viewer — a sandboxed <iframe> under a strict Content-Security-Policy. There is no separate hosting service to provision; publishing is just writing an .html file to the user's drive as the signed-in user, then opening the file's webUrl.

This skill teaches an agent two things:

1. Author HTML that actually renders — the OneUp CSP + sandbox fail silently, so non-compliant HTML shows up blank or half-broken with no error. Follow the contract below and it just works.
2. Publish it — upload the file with Microsoft Graph (primary) or the SharePoint Vroom API (_api/v2.0, for agents already holding a SharePoint token), then hand the user the returned webUrl.

The canonical, always-current copy of the authoring contract lives at your tenant's discovery endpoint: https://<your-sharepoint-host>/_html (human page), /_html/llms.txt (agent text), and /_html/help.json (structured). When in doubt, fetch /_html/llms.txt — it is generated from the live viewer source.

⚠️ Before you publish: you are writing to the user's real drive

Publishing with this skill creates or overwrites a file in the user's own OneDrive / SharePoint, authenticated as that user. It is not a throwaway sandbox.

• Nothing becomes public automatically. The file is private to the user until they explicitly share it. But it is a real document in their drive.
• You MUST get the user's explicit confirmation before every upload and every overwrite. Show them the exact target (drive, folder, filename) and whether it creates a new file or replaces an existing one, and wait for a yes. See Confirm before writing.
• Overwrite is destructive. PUT .../content to an existing path replaces the file's contents. Confirm the path and that overwrite is intended.

The authoring contract (make HTML render in OneUp)

Your HTML runs inside <iframe sandbox="allow-scripts"> under a strict CSP that fails silently. Follow these rules or the page renders blank/partial with no error.

Content-Security-Policy applied to your HTML:

default-src 'none';
script-src 'unsafe-inline' 'unsafe-eval';
style-src 'unsafe-inline';
img-src data: blob:;
font-src data: blob:;
connect-src 'none';
worker-src 'none';
object-src 'none';
base-uri 'none';
form-action 'none';

The wrapper iframe additionally applies frame-src 'none'.

Non-negotiable rules:

1. No external resources. No <script src>, <link href>, <img src=https://…>, @import, url(https://…), remote fonts, or CDNs. Inline everything or use data: / blob: URIs.
2. All script must be inline <script> blocks. No module imports, no import(), no <script src>.
3. All styling must be inline <style> or style="…". No <link rel="stylesheet">.
4. No network at runtime — fetch, XMLHttpRequest, WebSocket, sendBeacon, EventSource are all blocked (connect-src 'none'). To read SharePoint file data at view time, use the LiveData manifest (below) — it is the only sanctioned data path.
5. No storage primitives — localStorage, sessionStorage, indexedDB, document.cookie (opaque-origin sandbox). For cross-reload state, set location.hash directly (e.g. location.hash = '#slide=3'). Do not use history.pushState/ replaceState — the sandboxed document's base URL is inherited from the wrapper's blob: origin, so any url argument (even a bare #hash) throws SecurityError.
6. No top-level navigation, popups, form submission, workers, or plugins. alert / confirm / prompt are no-ops.
7. Inline on*= handlers work (they're rewritten to nonced scripts) but are an XSS sink if they template untrusted data — prefer addEventListener for computed/user/fetched data.
8. Save as UTF-8 (the file-creation APIs below write UTF-8; add a BOM if authoring by hand and non-ASCII renders wrong).

Links: <a href> opens only for same-tenant SharePoint hosts or the external allowlist; javascript: / vbscript: / data: hrefs are rejected.

Live SharePoint data (the only runtime data source)

Runtime fetch is blocked. To show live SharePoint file content, embed a LiveData manifest — the host prefetches each item server-side with the viewer's credentials and injects the results before your script runs:

<script type="application/json" class="ka-livedata-manifest">
[{ "spItemUrl": "https://contoso.sharepoint.com/_api/v2.0/drives/{driveId}/items/{itemId}", "fileName": "data.csv" }]
</script>

Before your script runs, the host sets window.__LD_RESULTS__, keyed by the exact spItemUrl. Each value is { format, content, error?, metadata } — read content (a string) and check error.

Non-obvious but essential:

• Always include fileName (with extension) in each item — the host picks the csv / xlsx-sheets / text decoder from it. Without it an .xlsx comes back unreadable and a .csv is not parsed as csv.
• spItemUrl must contain /drives/{driveId}/items/{itemId} (the origin is ignored, rebuilt from the viewer session). Get those ids by resolving the file: GET /shares/{u!token}/driveItem → parentReference.driveId + id.
• Refresh RE-LOADS your iframe with fresh data — postMessage({type:'ka-html-viewer-refresh'}, '*') (debounced 5s). There is no inbound "updated" event to listen for.
• Treat the content as untrusted — build DOM with textContent, never innerHTML it.
• Caps: 32 items/page (over-cap drops all results), 50 MB/item; formats: text, csv, xlsx-sheets, rows.

SharePoint lists — a manifest item can also point at a list (kind:"list" with webUrl + listId, plus optional select of INTERNAL column names). The host renders it server-side and injects it as format:"rows" — content is a JSON string { headers, rows } (already flattened; up to 100,000 rows, truncation sets a non-blocking warning). Resolve listId + internal column names at authoring time via GET {webUrl}/_api/web/lists/getbytitle('{Name}')/fields?$select=InternalName,Title,TypeAsString (or Graph /sites/{site-id}/lists/{list-id}/columns). Internal names can differ from the display title (e.g. "City" → Ciity). See references/live-data.md.

Full protocol + a complete worked dashboard: references/live-data.md (also mirrored in your tenant's /_html/llms.txt). Table-only starter: assets/templates/livedata-dashboard.html.

Don't have HTML yet? Start from a template

If the user wants to publish something but has no HTML, use a ready-made, CSP-compliant template from assets/templates/ instead of hand-writing:

Read the chosen file, replace its <!-- REPLACE: … --> placeholders with the user's real content, verify it against the contract above, then publish.

Step 1 — Author (or fill in) compliant HTML

Produce a single self-contained .html document that obeys the contract. Everything inline; no external references; LiveData manifest if it needs SharePoint data.

Step 2 — Get an access token

Publishing is authenticated as the user. You need an OAuth 2.0 bearer token (Azure AD).

Primary — Microsoft Graph:

• Scope (delegated): Files.ReadWrite (write to the user's own drive). Use Files.ReadWrite.All or Sites.ReadWrite.All to write to other libraries/sites the user can access.
• Acquire it with the user's identity via MSAL (interactive, device-code, or an existing signed-in session). Do not invent or hardcode tokens. If you don't already have a Graph token, run an MSAL flow — see references/api.md.

Alternative — SharePoint Vroom (use when the agent is already onboarded with a SharePoint OAuth token, e.g. an add-in / SPFx / on-behalf-of flow): a token whose audience is the SharePoint resource (https://{tenant}.sharepoint.com or -my.sharepoint.com), scope AllSites.Write / MyFiles.Write. Same driveItem calls, different base URL.

Full token flows, scopes, and consent are in references/api.md.

Step 3 — Confirm, then publish

Confirm before writing

First preflight the HTML — run scripts/preflight.sh <file>; it greps for forbidden APIs and checks LiveData manifest items (file items need spItemUrl + fileName; kind:"list" items need spItemUrl + webUrl + listId) ("renders blank with no error" is the top failure). Then list the destination folder (GET …/children) to detect a name collision, so you can tell the user accurately whether this is a create or an overwrite.

Before the upload call, show the user exactly what will happen and get a yes:

About to publish to SharePoint/OneDrive:
  Account:   user@contoso.com
  Location:  OneDrive  ›  Documents/Agent HTML/
  File:      sales-dashboard.html
  Action:    Create new file            (or: Overwrite existing file — REPLACES its contents)
  Renders at: <webUrl>?web=1   (?web=1 opens the OneUp viewer; the bare webUrl downloads)
Proceed?

Only continue on an explicit yes. For overwrite, make the destructive nature explicit.

Publish via Microsoft Graph (primary)

Simple upload (files ≤ 4 MB — plenty for inlined HTML). New file by path:

PUT https://graph.microsoft.com/v1.0/me/drive/root:/Agent HTML/<name>.html:/content
Authorization: Bearer <token>
Content-Type: text/html

<!DOCTYPE html> … your document …

Response 201 Created (new) / 200 OK (replace) → a driveItem. Read webUrl, then give the user <webUrl>?web=1 — the ?web=1 opens it in the OneUp viewer; the bare webUrl downloads the .html file:

{
  "id": "01ABC…",
  "name": "sales-dashboard.html",
  "size": 84213,
  "webUrl": "https://contoso-my.sharepoint.com/personal/user_contoso_com/Documents/Agent%20HTML/sales-dashboard.html",
  "file": { "mimeType": "text/html" }
}

For files > 4 MB use an upload session (createUploadSession) — see references/api.md.

Publish via SharePoint Vroom (alternative)

Same shape, SharePoint base URL, SharePoint-audience token. Use this when you already hold a SharePoint OAuth token (SPFx / add-in / on-behalf-of) or want to target a specific SharePoint host or site directly. (graph.microsoft.com already routes own-drive uploads to the caller's farm — including dogfood rings like *.sharepoint-df.com — so Graph alone is usually enough.)

PUT https://{tenant}-my.sharepoint.com/_api/v2.0/me/drive/root:/Agent HTML/<name>.html:/content
Authorization: Bearer <sharepoint-token>
Content-Type: text/html

<!DOCTYPE html> … your document …

Returns the same driveItem (read webUrl). Full details, including site-drive paths (/_api/v2.0/drives/{driveId}/items/{itemId}/content), in references/api.md.

Publish into a SharePoint site library (instead of your OneDrive)

Same call against a site drive. With Graph (scope Sites.ReadWrite.All):

PUT https://graph.microsoft.com/v1.0/sites/{site-id}/drive/root:/Agent HTML/<name>.html:/content

or Vroom against the site host:

PUT {siteUrl}/_api/v2.0/drive/root:/Agent HTML/<name>.html:/content

The site drive root is its default document library ("Shared Documents"), so keep the path root-relative. Resolve {site-id} with GET /sites/{host}:/sites/{path} (or /teams/{path}).

Publish by folder URL — fewest calls (Shares API)

If you only have the destination folder's URL, skip resolving drive/site ids — encode the URL as a sharing token and upload into it directly:

1. Encode: u! + unpadded base64url of the folder URL (base64, strip =, /→_, +→-).
2. GET /shares/{u!token}/driveItem → the folder's id + parentReference.driveId (one call).
3. PUT /drives/{driveId}/items/{folderId}:/<name>.html:/content → upload.

That's 2 calls, and it works for any OneDrive, site, or library. Details in references/api.md; ready-made script: scripts/publish-to-folder.sh.

Step 4 — Open it

Give the user <webUrl>?web=1 (the webUrl from the response with ?web=1 appended). The ?web=1 renders the HTML in the OneUp viewer; the bare webUrl downloads the .html file. webUrl is stable across updates, so links keep working after you modify.

Updating / modifying an asset you created

To change a published file, overwrite it — same path (or item id), same call:

PUT https://graph.microsoft.com/v1.0/me/drive/items/{item-id}/content
Authorization: Bearer <token>
Content-Type: text/html

<!DOCTYPE html> … updated document …

or by path: PUT /me/drive/root:/Agent HTML/<name>.html:/content.

Overwrite replaces the file's contents — confirm with the user first (show the target

• that it replaces the existing file). The webUrl stays the same. Vroom is identical with its base URL. To rename/move instead of replace content, use PATCH …/items/{item-id} (see references/api.md).

Error codes

Blank / half-rendered page after opening → an authoring-contract violation (usually an external resource or a runtime fetch). Re-check the contract.

Reference

• references/api.md — full Graph + Vroom endpoint reference: authentication & scopes, token acquisition, request/response schemas, upload sessions, metadata update, and the driveItem shape.
• references/contract.md — the OneUp authoring contract in full.
• assets/templates/ — ready-made CSP-compliant templates.
• Live docs (authoritative, tenant-specific): https://<your-sharepoint-host>/_html, /_html/llms.txt, /_html/help.json.