> ## Documentation Index > Fetch the complete documentation index at: https://docs.agentlinkops.com/llms.txt > Use this file to discover all available pages before exploring further. # Inspect internal links from a page list > Check supplied pages, understand coverage limits, inspect saved connection evidence, and prepare an internal-link task for your agent. Use **Internal links** to check how pages on your website connect. Supply a page list, run checks when you are ready, then inspect the saved HTML evidence before deciding what to change. The scan checks the URLs you supply. It does not discover every page on your website, render JavaScript, or edit the site. ## Supply a useful page list Choose your website, open **Internal links**, and select **Add pages**. In **Inventory**: 1. Choose **Where this list came from**. You can use your own list or page URLs extracted from a sitemap, CMS, repository, or search-data export. For a sitemap, paste its page URLs; the form does not fetch the sitemap file. 2. Paste up to **100 distinct page URLs**, one per line. Use full HTTP or HTTPS URLs with the website's exact hostname. Subdomains and a different `www` version need their own website. 3. Optionally add **Priority pages** that matter to you and up to **10 starting pages** for measuring link distance. Both must already appear in your page list. Starting pages do not cause the scan to discover more pages. 4. Select **This is the complete page inventory for this website** only if the list is complete. Leave it unchecked for a sample. AgentLinkOps records your declaration without verifying that the list contains every page. 5. Select **Save page list**. Saving creates a separate scan without fetching pages. Saving counts the supplied pages toward the workspace allowance of **300 pages per UTC day**, even before checks run. A website can retain up to **20 scans**, each for **30 days**. Start with the pages relevant to the question you want to answer. ## Check pages and keep the coverage in view Select **Check these pages**. The progress strip separates checked pages, pages that could not be checked, and pages still waiting. A finished scan can still contain unknown results. **Pause checks** stops after the current batch and keeps completed checks. Return to the **Saved scan** and choose **Continue page checks** to process waiting pages. Closing the page stops the app from starting further checks; return to the saved scan to continue. Checks can also pause for the website's check interval or after ten minutes. Follow any displayed wait time before continuing. This workflow does not set up recurring checks. **Refresh saved evidence** reloads saved results. It does not fetch the website again. To collect fresh evidence for pages already processed, including pages that could not be checked, save a new page list and run a new scan. Checks use raw HTML. Links added by JavaScript may be missing, and a blocked or incomplete capture does not establish that a link is absent. Even a list you declare complete still has these capture limits. ## Read findings as questions to investigate Open **Health** and review **Where to focus**. Expand **Finding evidence** for the recorded facts, then select **Inspect connections**. | Finding | What the saved evidence establishes | | - | - | | Orphan candidate | No incoming link was observed from the supplied inventory. This does not prove that the page has no incoming links elsewhere on the website. | | Priority page without contextual incoming links | A page you marked important has no observed contextual incoming connection. Its importance comes from your selection. | | No contextual outgoing links | None were observed in the page's complete static HTML. Review useful reader connections; there is no required link count. | | No route from starting pages | After all listed pages were completely checked, no route was found within that list. Pages outside the list may provide a route. | | Broken destination | An observed link points to a listed URL that returned HTTP 404 or 410. Check the source and destination observation dates before editing. | | Could not completely observe a page | Missing links remain unknown. Inspect the recorded reason before drawing conclusions. | No findings does not establish that the whole website is healthy. Check the inventory coverage and unknown pages first. ## Inspect the connection behind a finding In **Page connections**, choose a **Page to inspect**. The graph shows incoming and outgoing links, with up to six incoming and six outgoing pages from the currently loaded connections. Use **Next evidence** to read further results in the table; the graph is not a whole-site map. **Show contextual links only** is selected initially. Here, contextual means the link was found in an HTML main-content landmark. It does not establish reader relevance or rendered visibility. Clear the filter to include template and unclassified links. Open a connection row to see its link text, page region, retained surrounding passage, observation time, and capture hash. Use **Open source page** to inspect the current page before proposing a change. A linked destination outside your supplied list may appear, but its own check status and outgoing links remain unknown. Desktop Page connections view shows an example home page linking to its guide, with the recorded connection below. Example scan with two checked pages and one page that could not be checked. All three pages have been processed; the graph and table show one observed connection. [Open full-size page connections image](/images/internal-connections-synthetic-98edcb9b14.png) ## Give your agent evidence and a specific task Choose **Prepare an agent task for this page**, or **Copy task for your agent** from the page menu. The task includes the saved scan reference and asks the agent to follow evidence pagination, check current content and existing links, and propose a useful connection in an exact passage. If clipboard access fails, copy the displayed **Agent task text** manually. Review the proposal in your existing website workflow. Copying the task does not publish a change or schedule checks. Keep the scan and finding references with the proposal, and record any later authorized change separately. Use **Download scan evidence** to save a JSON file containing the scan, pages, connections, and findings. The download follows pagination and stops if the scan changes during export; refresh and retry in that case. Export evidence you need before its displayed retention expiry. See [sync and export](/guides/sync-and-export) for keeping other AgentLinkOps records in your local workflow. ## Run the same workflow through commands | Step | Command | | - | - | | Save a supplied inventory without fetching | [create\_internal\_link\_snapshot](/reference/commands/create_internal_link_snapshot) | | Process pending pages in small batches | [process\_internal\_link\_snapshot](/reference/commands/process_internal_link_snapshot) | | Read progress, coverage, and retention | [get\_internal\_link\_snapshot](/reference/commands/get_internal_link_snapshot) | | Find a retained scan for this website | [list\_internal\_link\_snapshots](/reference/commands/list_internal_link_snapshots) | | Read pages, connections, or findings | [query\_internal\_link\_snapshot](/reference/commands/query_internal_link_snapshot) | Reuse the creation request's `requestId` with the same payload after an uncertain response. Prefer `maxPages: 1` for interactive processing and respect returned retry timing. Follow query cursors using the same snapshot ID. Reading requires `watches:read` for the website. Saving and processing also require `watches:write` and a role allowed to manage links. The app's download action also requires `exports:create`. Current workspace membership and website access are checked when requests run. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.