Follow activity and improve document quality
Use Activity to understand what happened. Use Lint queue to find document issues and unanswered knowledge needs. These are useful even if you do not write code: “lint” means checking content against the enabled rules.
In v0.2.1, Lint queue has a Preview badge in the left navigation: this area is still being refined. You can open it there without a Labs setting. Inspect findings and suggestions before acting, and follow the permissions and actions described below.
Follow recent work
The Dashboard's Recent activity previews up to eight entries. Choose View all activity to open Activity with the same period and read-grouping choice. System events, such as sign-ins and background jobs, are kept out of this preview; find them in Activity with Type → System.
- Open Activity and check the workspace.
- Choose a Period and Type: Workspace, MCP, System, or All. Where available, use Sub-type to narrow that type, such as Context or Review under Workspace, or Jobs under System. On a phone, open Filters first.
- Read Pending, Advisory, and History.
- Open a linked title to inspect its target. A pending review card opens that exact item in Review; Review required is a status, not a second link.
Pending events may need review; advisory events need attention; history contains completed changes. These groups organize activity and do not replace a target's actual state. For example, check the Context published version before concluding that everyone can read a change.
Read an activity row
Dashboard and Activity use the same compact list, with thin dividers between entries. The document title leads the row; the section icon identifies the area, such as Context or Review. The action badge tells you what happened, such as Read, Edited, Published, or Failed; read that label before treating a tool call as a completed change.
The actor chip identifies who acted. A person icon marks a person and a cog marks the system. An MCP entry can show both the credential owner's name and the AI client's name; the client portion keeps its own color. Select a shortened name to read the full identity. With a keyboard, focus the chip, press the down arrow to enter the name panel, and Escape to close it.
The markers in the Context, Library and Agent explorers summarize only an item's latest content change within the last 72 hours and, for Editors and Owners, a review open on its current draft. Use Activity for the full history of events. See Read the explorer markers.
Group repeated reads is selected by default in Dashboard and Activity. Compatible repeated MCP reads can share one entry. Choose Expand … original events to inspect the individual reads and Collapse events to hide them, or clear the checkbox to list individual events throughout the feed. Other activity, including edits and reviews, remains separate.
A group joins consecutive successful reads by the same client in the same session, using the same tool, document version, and read options, within five minutes. Each group contains at most 50 events and stays within one recorded calendar date. A different section, version, or private editing state can keep reads separate. Grouping keeps the original records and respects your access; it does not share another author's private MCP edit. Older records without enough version evidence can remain ungrouped or have no document link. Expanded members use a short row for each event; the group shows its start and end times beside the relative time.
The count distinguishes displayed entries from the original events they represent. For example, 3 entries · 8 original events does not mean that only three actions occurred. Counts describe the Current query range; a + on a lane's count means more matching activity exists beyond that loaded range.
The overview shows a limited window of newest entries. Use View all pending activity, View all advisory activity, or View all history activity, then Load more to inspect older activity in that lane. All matching activity in this snapshot is shown means that lane's current query has been exhausted, not that no new activity can arrive.
Some records show No target link. Use View MCP activity or View activity list to inspect the matching activity instead. For MCP calls, Call details can show the tool, result, duration, and public error code; the record does not always identify a page you can open.
A background job can show progress, stage, or an error. A failed attempt with a scheduled retry differs from a completed conversion. If Library remains stuck, capture the job's visible error and ask the administrator.
Inspect quality findings
- Open Lint queue and choose Findings → Open.
- Filter by Layer or Severity if the list is long.
- Open the target and read the finding and Suggested next step.
- Correct the actual page or source where needed.
- An Editor or Owner can run checks and update the finding's status.
Viewers can read permitted findings but cannot run lint or resolve them.
| Action | What it does |
|---|---|
| Mark reviewed | Records that the finding was acknowledged. |
| Record as fixed | Changes the finding's status. It does not edit the page. |
| Ignore finding | Removes it from the unresolved queue and keeps a record in resolved history. |
Read the confirmation before a bulk action. Some selected findings can succeed while others fail; check the result rather than assuming all were updated.
Fill knowledge gaps
Choose View → Coverage gaps to inspect queries that did not find usable evidence. For example, repeated questions about the launch reviewer can reveal that the plan never names one.
Open the linked finding if present, locate the right source, and write or update a Context page. Search again, or ask again if you turned on the experimental Ask menu, to check whether the new material answers the question. A coverage gap identifies missing retrieval evidence; it is not an automatic instruction to invent an answer.
AI suggestions, when present, are proposals to inspect. They do not mean the content was automatically corrected or its factual accuracy certified.
Next: Context pages, Search and Ask, or Review.
Save and review policy
An Owner opens Settings → Content policies → Lint. Inspect Clean up on save, Review and publication, and the collapsed searchable Checks list. Read Current effect before changing an option. If loading fails, unknown controls stay disabled; retry options and checks independently. Required safeguards stay on, and external-source findings are read-only.
| Option | Effect |
|---|---|
| Auto-fix safe formatting on save | Cleans supported spacing and tables while preserving wording, code and hard breaks. Safe formatting-only changes can qualify for immediate approval when review safeguards pass, independently of the link-fix approval switch. |
| Fix internal links → Automatically | Converts supported readable internal links to stable document addresses. Published maintenance can also repair eligible links. |
| Fix internal links → Suggest only | Leaves ordinary link corrections as suggestions; it does not start published-document repairs. |
| Automatically approve safe link fixes | Allows safe fix-only changes to be approved immediately when review safeguards pass. A new reference to an unpublished destination still needs the appropriate review. |
Exact readable system index references become supported document links when authored content is saved, even with optional formatting off or ordinary links in suggestion mode. Restoring or copying history preserves that saved version's text.
A queued Lint run can repair eligible published Context links through the normal version, review and publication workflow. Another active edit, a pending review or a changed published version can prevent automatic repair. Citations and references are retained. If an integrity check fails, the run saves none of its repairs; inspect the resulting version and review status before retrying.
Fix an unresolved citation
Unresolved citation means a marker does not identify a usable document, for example an old @cite{key} or an incomplete citation. Open the page's References chooser, find the intended accessible document, and use Cite in text to replace the marker with its document address. Save, inspect the result, and publish when ready. A bibliography-only entry is different from an in-text marker; review both the page and Reference changes. See Images, links, and citations.
Read the new activity cards
The document name comes first, followed by what happened. The whole card opens one document or review; the time uses the same destination. Identity chips, grouped reads and Call details have separate controls. Review requested describes a request; Approved, Requested changes and Rejected describe decisions. Dashboard omits technical call fields and links to MCP Activity. There, Call details names the tool, result, duration, failure meaning and next step, with raw IDs/codes available as chips. A failed review request is not a completed review.
Read-together signals can be disabled by an Owner in Settings → Workspace → Knowledge graph. They describe consecutive reads in MCP conversations, at most 30 minutes apart, with supporting session counts and time ranges. Turning them off hides them from local graphs immediately and background cleanup removes the projection; underlying MCP read history is retained for rebuilding if enabled again. These signals are associations, not citations.
Implementation reference: Activity labels and filters, activity entries and links, full actor names, read grouping, read visibility and grouping conditions, Lint actions and status-only fixes, Preview menu, explorer markers.