Skip to content

Web publishing security and privacy

Static publishing is a trust boundary. The acquisition side handles Atlassian authentication; the public output must not inherit credentials, private source URLs, raw HTML, or runtime access to Confluence.

  • ADF/Storage is validated and normalized before Astro sees it.
  • Bundles and build inventories are private, digest-bound, and schema-validated.
  • The public output contains rendered content and verified asset bytes, not raw source bodies or credentials.
  • Strict completeness prevents an apparently complete site from hiding an unreadable page.

The render kit escapes text, rejects unsafe schemes and path traversal, and resolves page/asset links through trusted semantic keys. Active content such as iframes, scripts from content, credentialed URLs, and unsafe SVG is rejected or degraded to a visible safe fallback. The verifier crawls every owned output, link, fragment, image, and resource sink.

Analytics is disabled by default. The optional Plausible adapter sends only a pathname-only pageview to an allowlisted HTTPS /api/event endpoint with credentials: omit. It strips query strings, fragments, titles, source ids, Confluence URLs, search terms, and arbitrary properties; it has no queue, replay, or content cache and respects Do Not Track. Blocking the endpoint must not affect page rendering or search.

Operators remain responsible for lawful configuration, consent, retention, and their analytics provider’s terms.

An “Edit in Confluence” link is optional and off by default. See edit-link configuration for the fields.

  • Source of the URL: refresh takes the edit URL (Cloud editui, Data Center edit) or, with fallback: "open-page", the page view URL (webui) that Confluence returned for the page. It never builds a URL from a page ID, title, or page content.
  • Trusted origin: a URL is kept only when it is https, on the Confluence site of the profile that ran the refresh, and free of credentials, fragments, secret-bearing query parameters, and REST API paths. Anything else is dropped and the page is published without a link. A profile whose site is not https cannot enable edit links.
  • Scope: only pages in the bundle get a link. Excluded, out-of-scope, and unreadable pages, label landings, and the 404 page get none.
  • Tenant disclosure: an edit link reveals the address of your Confluence site. A project with visibility: "public" and edit-link visibility all must set publicTenantDisclosureAcknowledged: true; otherwise the project file is rejected. With edit-link visibility internal, a public site publishes no links and no site address.
  • Frozen policy: the links and the policy they were resolved under are part of the digest-bound bundle. build rejects a bundle refreshed under a different edit-link policy.
  • Verification: verify requires the link on exactly the pages the bundle holds one for, as an <a data-atlcli-edit-link> whose target is the frozen URL on the trusted origin. A link on any other page, to another origin, or in a build with edit links off fails verification. Confluence REST and api.atlassian.com URLs in markup stay forbidden everywhere.

The link carries data-pagefind-ignore, so its label is not indexed. Edit URLs are not written to SEO metadata, feeds, the sitemap, or analytics events.

The static page remains readable without JavaScript. The browser fixtures use a self-contained CSP with only the narrowly scoped wasm-unsafe-eval allowance needed by Pagefind’s WASM runtime; broad unsafe-eval is not permitted.

Jira and Confluence are trademarks of Atlassian Corporation Plc. atlcli is not affiliated with, endorsed by, or sponsored by Atlassian.