The Page Health Auditor (Rovo agent)

Page Health Auditor is the conversational half of StagBane- a Rovo agent you talk to in chat. The review queue tells you which pages are risky by their metadata; the agent reads whether a page's content still holds up, and can act on your confirmation. It is optional: the queue works without it, and you only need it (with Rovo enabled and credits) for the content-level work below.

Who can use it. Any licensed user can audit or compare specific pages they can already see. The actions that change state, and the site-wide "which pages are stale?" list, are gated- see Acting on your confirmation and Listing stale pages.

What it does

Capability What you ask Who
Audit a page "Is this onboarding doc still accurate?" anyone (pages you can see)
Compare two pages "Do these two policies contradict each other?" anyone
Find overlapping pages "What else might conflict with this page?" anyone
Act on your behalf "Mark this current", "ask the owner to review it" owner / site admin
List stale pages "Which of my pages are stale?" site admins

Reading a page

  • Audit fetches one page and judges it: likely current, likely stale, or cannot judge. Every verdict cites specific text and names the page version it judged, so two verdicts are always explainable as different versions.
  • Compare judges whether two specific pages contradict each other. Both must be pages you identify; the agent will not guess a counterpart.
  • Find overlapping pages searches for candidates that might conflict with a page you name. It returns candidates to compare, never a verdict- a vocabulary-adjacent page presented as a match is worse than finding nothing, so when unsure it says "No confident match."

The agent reads as you (not as the app): it sees only what your own account can already open. Name a page you cannot access and it says so, rather than reading it for you.

Grounding: it speaks with the queue's voice

When a page is in your audited index, the agent is handed the queue's own deterministic finding- its band (Critical / Stale / Watch / Healthy) and the reasons behind it- and cites it, then adds its own read of the content and explains where the two diverge. Its highest-value move is exactly that gap: a page banded Healthy on a fresh timestamp whose content is plainly out of date, or a page banded Stale only because one link broke while the substance is fine. The agent never changes a band; the queue stays the product's verdict.

Acting on your confirmation

The agent can trigger the same non-destructive actions the queue offers- it never edits, archives, rewrites, or deletes page content, and it never stores its own opinion. It always states what it will do and on which page, and waits for your explicit yes before acting.

Action Effect Who can ask
Mark Verified Current Applies the verified-current label, clears the stale flag, resets the review timer the page owner or a site admin
Mark Stale Records a human "this is stale" verdict that bands the page Stale until it is verified again the page owner or a site admin
Request Owner Review Posts one comment @mentioning the owner (respects the daily request budget; never duplicates an open request) site admins
Set / clear needs-review Toggles the private needs-review worklist label (notifies no one; never affects the band) site admins

Permission is checked by the system on your identity, not on what the chat says- if you are not allowed, the agent tells you and points you to the queue, and never implies it acted when it did not. Page content can never authorize an action; only you can, in chat.

Listing stale pages (site admins)

Ask "which pages are stale?" and, if you are a site admin, the agent returns the queue's own ranked list (the most stale pages first, each with its band and reason). If you are not a site admin, it points you to the review queue rather than listing the site- the full ranked list lives there, gated to admins. The agent never assembles a site-wide list from search results; the queue's ranking is the only source.

Auditing several pages

Ask to "audit them all" (for example, after the stale list) and the agent audits up to five pages per reply- a short cited verdict for each- then offers to continue with the next five. Prefer one at a time? Say so, and it audits one page per reply and waits for your "next". The cap keeps a single reply's Rovo-credit cost predictable.

What it will not do

  • Change your content. It never edits, archives, or deletes a page. Every state change is one of the queue's non-destructive actions, and only on your confirmation.
  • Store its verdict. Audit and compare verdicts live only in the chat that produced them; they are never written back to the queue. (That is why each one cites a page version.)
  • See what you cannot. Reads run as you; restricted pages you lack access to stay invisible to it.
  • Crawl or rank the site itself. Site-wide ranking comes only from the admin list above or the queue, never from the agent stitching search results together.

Example prompts

  • "Audit https://.../wiki/.../pages/12345/Onboarding and tell me if it is still accurate."
  • "Compare the US and EU data-retention pages- do they contradict each other?"
  • "What other pages might overlap with our incident-response runbook?"
  • "Which pages are stale?" then "audit the first five."
  • "Mark page 12345 verified current." (the agent confirms first)
  • "Ask the owner of page 12345 to review it."

Availability and cost

The agent runs on your Rovo credits (about 10 per request), and needs Rovo enabled on your site. The review queue, the bands, and all queue actions work without Rovo- the agent is the on-demand content layer on top. See the user guide for the queue, and privacy & data for exactly what the agent reads and writes.