Tools
What the workspace agent can look up, change, and write. Ask for the job. It picks the tool.
You do not call these. Ask for the outcome. The agent picks the move, from the active crawl version unless a compare workspace is attached.
Pages and fields
| Capability | What it does |
|---|---|
| Find pages | Match URLs by path, title, status, or any crawl field |
| Open a page | The extracted record for one URL or a small set |
| Count and group | Breakdowns: status by folder, word count by cluster |
| Field map | Which fields exist on this crawl, and how complete they are |
| Structured query | A read-only query when a filter plus a group-by is not enough |
Meaning
| Capability | What it does |
|---|---|
| Similar pages | Pages or clusters close in meaning to a phrase or a URL |
| How related | How close two URLs are in meaning |
| Clusters | Every topic group, or the members of one |
| How the site hangs together | Site shape, near-duplicates, and hub pages the map leaves out on purpose |
| Concepts | Short topic phrases rolled up across the site or a cluster |
Issues and links
Findings by pillar, by rule, or by the pages in the worst shape. Severity sorts first; checks lead signals when the severity matches. Same inventory as the issues panel.
Ask about links across the site (hubs, orphans, depth), one URL's inbound and outbound links, or pairs in the same cluster that do not link yet. Ask for editorial or in-body links when navigation and footer links should not count.
Evidence
| Capability | What it does |
|---|---|
| Proof on the page | Tables, FAQs, lists, quotes, and JSON-LD the crawl stored |
| Open in the inspector | Puts that URL in the inspector so you can see the same tabs |
| Fetch a live URL | Reads a page that is not in the crawl, or re-checks one that is. Not crawl data. Treat it as fetched just now. |
| Live markup | The HTML for one region when the wrapping is the question |
When a reply includes Sources, each pill is an exit back to the evidence. Open one and Silkra applies that source as a crawl-table filter instead of leaving the citation trapped in chat.
The table and the workspace
Ask to show missing titles and filters land as chips. Ask to hide a class of pages for good and an Update exclusions card shows the proposed rule, then waits for Run.
After a sampled qualitative review, the agent may offer Save site trait. That writes the observation into the workspace context for later threads. It also waits for Run.
Between versions
What moved on the last refresh, the history of refreshes, and the body diff for one page.
Connected data
Search Console can return live clicks, impressions, CTR, position, and search queries for the bound property.
| Connection | What the agent can add |
|---|---|
| Ahrefs | Backlinks, organic keywords, and Site Explorer metrics |
| DataForSEO | Organic keywords, estimated traffic, and backlink summaries |
| Semrush | Domain, keyword, competitor, and backlink reports |
The connection must be on and available to the current workspace. Provider calls use that account's units or credits.
Connected actions
Webflow and WordPress can apply approved SEO title and meta-description edits. Webflow leaves a draft; WordPress keeps the page's existing publish state.
Jira, monday.com, and Linear can create a work item in the destination bound to this workspace.
Both moves pause on an approval card. Read the destination and proposed change before approving it. No external write happens first.
Compare workspace
Topic overlap, gaps, and strengths against the attached crawl. Paid. Attach the other workspace first.
Create artifacts
Tables, documents, and charts. Tables and charts need a paid plan.