Documentation

Anchor Audit user guide

How to use the app and understand its limits.

Product overview · Support

Read-only section link checks for Confluence Cloud. It checks internal links to headings and named anchors on the same Confluence Cloud site, including old Data Center fragments left behind by a migration. It suggests a fragment only when one target matches. It does not edit pages, anchors or links.

Open the app from either entry point:

  • Anchor Audit in the space navigation.
  • Audit section links in the page or blog actions menu.

Installation and access

The release installation path is through the Atlassian Marketplace. A Confluence Cloud site admin installs Anchor Audit on the intended site and reviews the access request. Pricing covers the site's licensed Confluence users, not only people who open Anchor Audit. Trials and billing are managed by Atlassian.

Open Anchor Audit in a space's navigation, or Audit section links in a page's More actions (•••) menu. The menu entry also appears on blogs; V1 scans page trees and reads blogs as link targets, so starting a scan from a blog shows a message asking you to open Anchor Audit from a page or space, with Close to leave the modal. It does not scan the blog.

On the first content read, Forge can ask you to Allow access. Approve the dialog to let the app read with your current Confluence permissions. If the app still shows Allow access after approval, select it to retry the read. As the scan restarts, the notice disappears and Pause and Abort become available. Declining access does not grant the app any additional rights. No separate external account is needed.

The app requests read access to pages (including blog targets), spaces and page trees. It reads with your current permissions and does not request write access. It saves only aggregate monthly usage totals, with no external service requests. Being a site or space admin does not make a restricted page visible unless you can already view it in Confluence.

Run a scan

  1. In a space, choose Scan whole space or Scan below a page. For a page tree, paste a link to a page on this Confluence site, or enter a numeric page ID. Links from the address bar, Copy link, and the editor are supported, including space home-page links and draft editor addresses. Draft content is not scanned. A blog address, with or without a date in the path, shows Choose a page to scan. Select Back to enter a page link or scan the whole space instead. The page action uses the current page and its descendants, with no page input.
  2. Select Start scan. The app reads current pages in chunks of at most 50, then reads any additional visible page or blog targets needed for that chunk.
  3. Review the summary and result table: source page, link text, target, status, cause and an optional suggestion. Problems focuses on unresolved results; All includes resolving links. Repeated links are grouped with occurrence counts.
  4. Use Pause and Continue within the same tab. After 500 pages, the scan pauses for your confirmation before reading more pages.
  5. Use Abort while running or Stop while paused to stop future calls. An already-running read can finish, but its late reply is discarded; committed results remain available in the tab. Stopping scan disables New scan until that request finishes. Then select New scan to discard the cache and read current content again.

Suggestions show the plain fragment and its percent-encoded form. Update a link manually in Confluence only after checking the target. An invalid anchor name with spaces also needs a manual anchor rename; changing the fragment alone is insufficient. The app has no repair button or file export.

Results exist in memory in this tab, with no saved scan history. Closing or reloading the view loses the scan; it cannot resume in another tab or session.

Understand the results

The summary counts link occurrences; the table groups repeated links from each source. Title-based links with different visible link text have separate rows, even when their unavailable targets share the same safe base. Occurrence counts account for repeated links within each group. The table shows 20 rows at a time. Use Previous page and Next page to review the remaining rows; Page 1 of 2, for example, shows your position. Sort results orders the full filtered list by source page or status before splitting it into pages. Switching between Problems and All, or changing the sort order, returns to the first page. The summary always counts the full scan. Unavailable targets show only a safe base path or ID, without a target title, slug, fragment or suggestion. Visible link text from the source page is retained.

StatusMeaningYour next step
ResolvesThe fragment matches a known heading or valid anchor, or the page-top rule.No link change is suggested. A duplicate-heading or collapsed-expand hint can still need review.
BrokenA fragment is empty, missing or uses a recognisable incompatible format.Review the cause. Use a suggestion only if the app found exactly one match.
AmbiguousSeveral possible headings or anchors match.Inspect the target and choose manually; no suggestion is supplied.
Target not availableThe target was not returned, was not found or is not viewable by you.Check your access in Confluence. This status does not establish that a page was deleted.
Can't verifyIncluded, legacy, synced, malformed or oversized content, or a target that takes too long to respond, prevents reliable checking.Inspect the target manually or start a new scan to try again. This is neither a confirmed broken link nor a verified match.
Not checkedA link or subtree falls outside the supported scope or exceeds a size limit.Review the omission count; a completed scan does not certify unchecked content.

Recognised causes include old Data Center heading or anchor formats, spaces in an anchor name, spaces instead of dashes, Unicode normalisation differences, case differences, an empty fragment and a missing heading or anchor. A link ending in # is reported as Broken / empty-fragment even though a browser may show the page top. Heading IDs and named anchor IDs are case-sensitive.

Progress, pauses and errors

Slow target lookups are retried in smaller groups. If an individual target still takes too long to respond, its links are marked Can't verify with the note “Confluence took too long to respond, so this link couldn’t be checked. Start a new scan to check it again.” The scan continues with the remaining targets and pages. Repeated links to that target keep the same result in this scan; start a new scan to try it again. If two individual targets time out without a successful target lookup between them, the scan stops with Scan interrupted and Try again. It keeps the results already shown; Try again resumes the unfinished checks. A new scan also rechecks any target previously marked Can't verify.

While linked pages are being checked, progress shows the source pages already read and the pages whose section links still need checking. Pages count as checked only after their section link results are ready. The progress bar appears while scanning and when the scan completes. It is hidden during pauses, while counts and results remain visible.

Old server links such as viewpage.action?spaceKey=DOC&title=Guide#Summary, without a page ID, are Not checked / Unsupported internal link. An explicitly invalid or repeated page ID is Not checked / Invalid URL.

StateMeaning and next action
Ready / idleChoose a scope and Start scan.
RunningA bounded read is in progress. Pause to wait, or Abort to stop future reads.
Paused at 500 pagesContinue checks up to 500 more pages; Stop keeps only results so far.
Paused by youContinue resumes in this tab; Stop ends the scan. A pending 429 preserves your pause, with Continue disabled until the wait ends.
Rate limitedConfluence returned 429. Wait for Retry-After (60 seconds if unavailable). The first retry is automatic; another rate limit requires Continue after the wait.
Near API capacityContinue anyway resumes with at least ten seconds between calls. Stop ends the scan; you can start a new one later.
FailedNothing was changed. Try again resumes from where the scan was interrupted. Slow source reads get one retry with fewer pages; slow page, blog and title lookups use progressively smaller groups. Two individual target timeouts without a successful target lookup between them stop the scan.
Stopping scanA stopped scan is waiting for its pending read. Its reply will be discarded; New scan stays disabled until it finishes.
Scan stoppedResults so far are shown. Use New scan for another run.
CompletedAll supported pages returned for this scope were processed. Review Not checked and Can't verify before drawing conclusions.
License inactiveAsk a site admin to start a trial or renew through Atlassian. All data operations enforce the backend license.
Page or space unavailableReturn to a page or space you can view; the app cannot expand your access.
Choose a page to scanThe entered link opens a blog post. Go Back and enter a page link instead, or scan the whole space.
No pages to check hereThere are no visible current pages in this scope. New scan returns to scope selection.
No section links found in the scanned pagesNo supported fragment links were found. Scan another scope returns to scope selection.
Couldn’t check linked pagesThe linked pages could not be checked. Try again resumes the scan; the chosen scope is retained.
Opened from a blogOpen the app from a page or space. Close leaves the modal without a scan.
The section link resolves / All section links resolveThe message gives the number of section links checked and says No problems found. Omitted links, uncertain content and expand behaviour still need their own review.

A 429 during your pause keeps the scan paused. The countdown enables Continue after the wait, and the scan resumes only when you select it. Stop remains available while waiting. The first automatic retry applies only to a rate-limit pause reached while running, and it happens at most once per scan.

Scope and limits

  • Confluence Cloud only; source scans cover current pages, not blog collections, archived content, drafts, comments, attachments or non-page content.
  • Only same-site internal links containing # are checked. External sites, other Confluence sites, short links (/wiki/x/…), unknown paths and comment or attachment targets are skipped. Links without a fragment are ignored.
  • A page-tree scan includes its root. Readable page roots at depth ten are continued and deduplicated. Pages below folders, whiteboards, databases or embeds at that boundary are counted as unchecked subtrees rather than silently certified.
  • Rate limits can pause a partial scan. There is no guaranteed whole-space completion time, scheduled scan or background continuation after closing the view.
  • Each resolver has a 20-second budget. Response-size limits can reduce a chunk, skip URLs over 2,048 characters or omit a single page's oversized link list. Slow page, blog and title lookups use smaller groups of targets before retrying. A large target index can become Can't verify. For an oversized target batch, the resolver compacts the largest target indexes first and stops as soon as the reply fits. Links to compacted targets are Can't verify, except empty and #top fragments. Empty fragments are Broken / empty-fragment; #top resolves to the page top (case-insensitive). Other targets remain checkable; compacted indexes are not reloaded automatically. Review them manually.
  • Renamed page titles, included/legacy content, anchors in macro bodies and unknown editor formats may prevent a suggestion. Check uncertain results manually.
  • Empty page titles or space keys in old-style page links are counted as Not checked, without stopping the other checks. IDs with leading zeros are normalized; invalid or oversized IDs are Not checked. Short links with section fragments in page content are still not checked; accepting a Copy link address to choose the scan root does not change that limit.

Privacy and removal

Forge reads visible page and blog content, titles, IDs and links to build section indexes. The authenticated browser receives indexes, source links, short link text and technical scan counts; paragraph bodies are not returned to the frontend. These values can include personal data already present in Confluence.

The app makes no external service requests and uses no external AI or analytics. It persists no scan data or settings in Forge storage, browser storage or a vendor database. The code emits no customer content to application logs. Atlassian manages its platform diagnostics, licensing and billing separately. Closing the view clears the app session; uninstalling does not delete your Confluence pages. Only aggregate monthly usage totals and a short coordination record are saved in site-scoped Forge storage. They contain no account IDs, page IDs, titles, query text or results. Monthly totals are replaced when work starts in a new UTC month. Technical records follow the host Confluence location and Atlassian's 28-day hosted-storage retention after uninstall.

Support

Use the support channel identified in the published Marketplace listing. Include the app version, status/cause code, page/link counts and a reproduction with fictional content. Do not send page bodies, titles, URLs, fragments, account IDs or unredacted screenshots. For security reports, use the security contact in the Marketplace listing.

Monthly usage allowance

Each site shares 20 processing batches and up to 4,000 requested source pages or link targets per calendar month (UTC). One audit can use several batches; retries and title lookups count again. The app stops before a request would exceed either allowance. Opening the app to view results already in the tab remains available.

The allowance also covers up to 120 KB of reads and 60 KB of writes for usage totals. If new work is temporarily paused, existing results remain readable in the current tab; closing or reloading still discards them. The app shows a clear message when work cannot continue. The monthly allowance can stop a scan before its usual per-run limit.