docs: sync docs with recently shipped Cube Cloud features

[keydunov]

Check List

• Tests have been run in packages where changes have been made if available
• Linter has been run for changed code
• Tests for the changes have been added if not covered yet
• Docs have been added / updated if required

Description of Changes Made

Routine sweep cross-checking recent cubejs-enterprise and cube-js/cube changes against docs-mintlify, filtered through the customer-facing criteria in cubejs-enterprise/.claude/shared/customer-facing-criteria.md. Found five small doc gaps (one stale/incorrect, four missing) and fixed them all here:

• docs/explore-analyze/workbooks/python-analysis.mdx — the "On dashboards" section said Python reports never re-run on dashboard load. That's no longer accurate: a dashboard filter or time-grain change now re-runs the Python analysis against the filtered SQL, on both the dashboard builder and published/viewing surfaces (not yet on anonymous/embedded-app dashboards). Corrected the claim and added the exception.
• admin/users-and-permissions/user-groups.mdx + roles-and-permissions.mdx — a user group can now be assigned a role (including Admin), conferring it on every member. Added a "Group roles" section and a cross-reference.
• admin/connect-to-data/multiple-data-sources.mdx — said the default data source "can be edited but not deleted." It can now be deleted from the UI like any named source (only the demo source can't be). Corrected the section.
• docs/explore-analyze/dashboards/widgets/charts.mdx — chart widgets on published dashboards gained an Edit in Workbook shortcut in their ⋮ menu (visible to users with workbook-edit access). Added a sentence documenting it.
• docs/explore-analyze/dashboards/dashboard-agent.mdx — for users with workbook-edit access, the Dashboard Agent on a published dashboard now has the same authoring capability as the Workbook Agent (data model edits, reports, drill-downs, publish, skills/sub-agents) — not just filter/time-grain changes. Added a "For editors" section and updated the can/can't list and the two-surfaces note accordingly.

No code changes; docs only.

---

🤖 Generated with Claude Code

https://claude.ai/code/session_01VjvAuTt42TokPntg4qSykj

---

Generated by Claude Code

[claude]

Claude finished @keydunov's task in 1m 29s —— View job

---

Docs review — approve with 2 medium consistency issues (0 high, 2 medium, 3 low)

Full review

• Read PR diff
• Check link refs / anchors resolve
• Check internal consistency of each edited page
• Post review

Docs-only PR; no code, tests, security, or performance surface. Review focused on factual consistency within the edited pages, link/anchor integrity, and Mintlify conventions.

Verified clean

• All new reference links resolve to real pages registered in docs.json: [ref-user-groups] → admin/users-and-permissions/user-groups, [ref-roles] → admin/users-and-permissions/roles-and-permissions (both at docs.json:282/:284).
• New anchor #for-editors matches the ## For editors heading; the "below"/"above" directional wording matches actual section order in the file.
• [ref-config-ref-env] in multiple-data-sources.mdx is still referenced elsewhere in the file (lines 85, 101), so removing its use in the "Editing and deleting" section doesn't orphan the definition.
• The multiple-data-sources.mdx rewrite is a net improvement — it replaces a stale "edit by hand" workaround with what actually happens on delete, and scopes the blast radius (own connection vars only).
• charts.mdx Edit in Workbook addition is well-scoped and correctly notes the permission gate.

Medium

1. python-analysis.mdx — the new filtered-re-run exception contradicts the unchanged sentence at line 100–101 ("viewing a dashboard never re-runs anything on its own"), which is exactly the line a reader will quote about compute cost. Also unstated: whether the filtered result is persisted or session-only. Inline comment
2. dashboard-agent.mdx — **It can't** (unless you're an editor…) applies the caveat to all four bullets including "Save its filter or time-granularity changes to the dashboard", but the For editors section never claims that capability, and Limitations still states unconditionally that filter changes are never saved. Inline comment

Low / nits

3. user-groups.mdx still has no trailing newline (pre-existing, but the PR touches the final line — cheap to fix while here).
4. user-groups.mdx frontmatter description still describes the page as only about data access policies; it now also documents role assignment. Worth a few words.
5. dashboard-agent.mdx — the "two different agent surfaces" bullet list at the top still describes the Dashboard Agent as viewer-only Q&A + filters. The paragraph below it now carries the editor exception, so it's not wrong, but the bullet is the part people skim.

Claims I could not verify

The behavioral assertions (demo source is the only undeletable one; group Admin assignment prompts for confirmation; editor-mode Dashboard Agent gets Workbook Agent tools; the embed/embedded-app gap for Python re-runs) originate from cubejs-enterprise, which isn't in this repo. Flagging as author-verified rather than reviewer-verified.

No prior review threads existed on this PR, so nothing to resolve or dedupe against.

• [Branch `claude/gallant-ramanujan-hgtxbv`](https://github.com/cube-js/cube/tree/claude/gallant-ramanujan-hgtxbv)

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
cubed3	🟢 Ready	View Preview	Jul 29, 2026, 12:14 AM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[claude]

This new exception contradicts the still-unchanged claim in the Running and refreshing section just above (lines 100–101):

Run is the only way the saved result changes. Opening the report, reloading the page, or viewing a dashboard never re-runs anything on its own.

With this change, viewing a filtered dashboard does re-run the analysis. The two statements are reconcilable (the filtered re-run doesn't change the saved result), but a reader hitting line 101 first will take away the wrong model — and it's the sentence most likely to be quoted about compute cost. Worth qualifying it there, e.g. "…or viewing an unfiltered dashboard never re-runs anything on its own — see On dashboards".

Also: the added text says the re-run "shows that result instead of the saved output" but doesn't say whether the filtered result is persisted. Given the surrounding page hammers "Run is the only way the saved result changes", one clause confirming the filtered re-run is session-only and doesn't overwrite the saved output would close the loop.

Fix this →

[claude]

The blanket "(unless you're an editor)" caveat now applies to all four bullets, but the first one is a different kind of claim from the rest:

• Save its filter or time-granularity changes to the dashboard — they apply to your session only and never change what other viewers see

The For editors section lists the added capabilities as "edit the data model, create reports and drill-downs, publish, and use skills and sub-agents" — it does not say an editor's agent can persist filter/time-granularity changes to the saved dashboard. As written, the caveat implies it can. That's also in direct tension with the Limitations section, which still states unconditionally that a filter/granularity change "applies to your session only — it is never saved to the dashboard."

Safer to scope the caveat to the authoring bullets and leave the filter one absolute, e.g. keep **It can't:** and append (unless you're an editor — see [For editors](#for-editors)) to the three authoring bullets, or split the list into "never" vs. "not unless you're an editor".

Fix this →