Skip to content

Deep scan

51 help pages

Deep scan lists the checks beyond Composer runs for one enrolled site, in three groups: Security, Maintenance and Upgrades. It sits between Deployments and Dependabot under Patching sections (address ?tab=patching&view=deep-scan). It appears only once Sentinel has turned it on for your organisation. Until then the tab is not shown, and an old link opens Runs instead. Any member who can open the site can read it. Only the security scan (below) can start anything from this page. Everything else on it is read-only, with links.

Each row shows:

  • A status badge. A security row with nothing open says No open findings. That means nothing was found by this check, not that the site is secure.
  • How fixes happen: Autopilot, Copilot or Report only. The dependencies row says Autopilot when its rule is on and the site is on Autopilot, and Copilot only otherwise (see Rules for this site below).
  • Included when the check costs nothing extra.
  • When it was last checked.
  • A link to where you act.

Today there are up to nine rows. The first five are the security scan's and sit first in Security, each only where Sentinel has turned the security scan on for your organisation:

  • Dependencies in every lock file: Composer, npm, Yarn and pnpm packages matched against known vulnerabilities.
  • Container images and their set-up: PHP and Nginx images, Dockerfiles and platform files.
  • CI workflows: GitHub Actions, for pinned actions and token permissions.
  • Custom code and Drupal settings: your own modules and themes, exported configuration and settings files.
  • Committed secrets: keys and passwords in the code and its recent history.

Each of these rows shows its state and the time of the last scan that answered it. The states are Checking (a scan that includes this check is running), Not checked yet (no scan has answered it), Out of date (a later scan did not answer it in full, or earlier findings are still open without a fresh answer), N open (the number of open findings), No open findings, and Needs set-up when the site has no repository (the link is Choose the repository). A scan answers a row only when every tool behind that check finished in full, so a failed or partial scan never turns a row to No open findings. No open findings means this check found nothing, not that the site is secure. Findings from a repository or branch the site used before do not count on the rows. These rows say Report only: they find problems but do not fix them, and they are Included. Each one has a See findings link that opens the check's own page (below).

A check's page. See findings opens the check inside the same view (address ?tab=patching&view=deep-scan&check= followed by the check's name). Any member who can open the site can read it. It shows:

  • A link back to Deep scan, the check's title, its state, Report only, and "Checked" with the time, branch and commit of the last scan that answered the check.
  • A note on what the check reads. Sentinel reads the repository only, so an empty list never means the site is secure.
  • The findings table, with severity, the finding (title and where it is), who found it, when it was first seen and its status. It shows open findings first, most severe first, 50 to a page with Previous and Next. Tick Show fixed and dismissed to see every status. Only findings from the repository and branch the site uses now are listed, the same ones the row counts. When the list is empty it says "No open findings from the last scan on" that date, or "No scan has answered this check yet."
  • Downloads: CSV and SARIF of the list as you see it (open findings only, or every status), and Download software list (CycloneDX) on the dependencies check only.
  • Scan history: the scans that ran this check, each with its date, what it said about the check (Checked in full, Not every tool ran in full, Did not finish, Cancelled, Waiting to start or Running), how it started (for example Scheduled or Requested), the branch and the commit.

A finding. Select a finding's title to open its panel: a summary, where it is, the file, what version fixes it, who found it, the rule, the CVSS and EPSS scores, links to NVD, GitHub advisories, drupal.org, OSV and CWE, Is it reached, a How to fix box, The AI review says, when it was first and last seen, and the decisions made on it with their dates. For a secret the panel shows only its first and last characters and its length, never the value. A finding closes by itself when a later scan no longer sees it.

Fixing a finding. The How to fix box on an open dependency finding depends on where the fix is:

  • Fix with Patching. For a Composer package in the lock file next to the site's composer.json, found on the repository the site uses now, with a fixed release on the line it is installed on, the box shows a Fix with Patching button. Organisation administrators and site managers (people who can run Patching) see the button. Everyone else sees that one of them can do it. The button queues an ordinary security update: a security pull request that moves the package to its fixed patch release, together with any other security updates the site needs. The panel then says "A security update is queued" with a Follow the update link. The usual reasons an update cannot run apply, and the panel says which: an update is already queued or running, a Sentinel security pull request is already open, the site has no repository, Patching is stopped for now, or the finding changed since the page loaded (a later scan may have closed it). The update needs the site to be enrolled and its tests to be adequate, and counts against your plan's job allowance like any update.
  • Fix in progress. While a Sentinel pull request moves the package, the finding shows Fix in progress in the table. The panel links to the pull request. The finding closes when a later scan of the merged branch no longer sees it, not when the pull request merges.
  • Only on a newer release line. The panel says Patching applies patch releases only, so upgrade the package by hand or with a maintenance update.
  • npm packages. For an npm package that the project names itself (in package-lock.json or npm-shrinkwrap.json), the box shows an npm install command with a Copy button. Run it in the folder of that lock file. For Yarn, pnpm and packages that only another package pulls in, the box says to update the package with your package manager.

Rules for this site. The dependencies check's page has a Rules for this site panel above the findings. It lets a finding start a security update by itself, so you do not have to press Fix with Patching.

  • Start a security update by itself. Off by default. Tick it, choose Only for this severity or worse (Critical, High (the default), Medium or Low) and select Save rules.
  • Only new findings. A finding first seen before you saved the rule starts nothing, so there is no backlog. Turning the rule on, or lowering the severity, covers findings from then on. Use Fix with Patching for older ones.
  • What starts an update. An open finding on the repository and branch the site uses now, for a Composer package in the lock file next to the site's composer.json, with a fixed patch release on the line it is installed on, at or above the chosen severity. Never a fix that exists only on a newer line, an npm package, another lock file, or a code or secret finding. The site's security schedule start time applies, for example a release window. Each finding starts at most one update, and several new findings share one.
  • What the update does. It is the same security update as Fix with Patching. On a Copilot site it opens a pull request and stops. On an Autopilot site (not "I deploy it myself") Autopilot merges, deploys and verifies under its usual rules, which include patch releases only and passing checks. Anything else stops at a pull request.
  • If you change the repository, the rule turns off until you save it again.
  • Needs security updates turned on for the site. Otherwise the panel says so and the box cannot be ticked.
  • Who. Organisation administrators and site managers (people who can run Patching) see the box and Save rules. Everyone else reads the rule as text: "New findings start a security update by itself, high or worse" or "New findings start nothing by themselves". On an Autopilot site, turning the rule on or lowering the severity asks for a recent second-factor code if you have set one up.
  • Where you see it. The Activity log says "New security scan findings now start a security update" or "New security scan findings no longer start a security update", and a run started this way says "Started by a security scan finding". The check page's badge and the dependencies row say Autopilot or Copilot only.

Where the security scan is not on for your organisation, none of this appears.

Deciding on a finding. Organisation administrators and site managers (people who can run Patching) see three buttons on an open finding: Mark false positive (the tool is wrong), Dismiss (it does not matter for this site) and Accept risk. Each needs a reason of 3 to 2000 characters. Accept risk also takes an optional Opens again on date, at most a year away. Leave it empty for the usual period. The finding opens again on that date. Reopen ends a decision that is still active. Viewers, members and other readers see the panel without these buttons.

Scan selected. Organisation administrators and site managers (people who can run Patching) see a checkbox on each of the five rows and a Scan selected button at the top of Security. Tick the checks you want and select the button: the scan runs only those checks, and the message "Scan queued. The rows update when it finishes." appears. Every box starts ticked, and your choice is kept only while you stay on that site. Committed secrets is always ticked and cannot be unticked, because it always runs ("Always runs: what it finds is hidden from the other checks' output."). The button is off, with the reason under it, when no other check is ticked ("Tick at least one check.") or when the site cannot be scanned now: a scan is already running, your plan's jobs are all busy, more sites are enrolled than your plan covers, there is no active subscription, the site is not enrolled or is switched off, the repository is missing or its connection needs attention, an AI provider key is missing or was revoked, or the AI terms or data processing agreement are not accepted. If the request is refused when you press it, the same reasons are shown in place of the queued message. A scan counts against your plan's Patching job allowance like any job. Viewers, members and any other reader see the rows without checkboxes and without the button.

Security scan alerts. Where the security scan is on for your organisation, Sentinel tells you when it finds something serious, so you do not have to open Deep scan to notice. Everything below is in Messages and the message bell, under the Patching category (see Messages and alerts).

  • New findings. After each security scan, and after the daily check of your sites against Sentinel's list of known vulnerabilities, a site that has new or reopened findings gets one message, not one per finding. It is titled "(site) has new security findings to review" (a warning), or "(site) has new critical security findings" (critical) when any of them is critical. The message says how many findings are new or reopened and lists up to ten, worst first, with where each one is, its rating and whether it is new or reopened. When there are more, it adds how many more. It ends with Open Deep scan, which opens that site's Deep scan view.
  • What counts. Findings rated high make a warning. Findings rated critical, findings on the CISA Known Exploited list, and anything that looks like a credential in your repository make a critical message, whatever its rating. Findings below high never send a message. A finding the scan's review decided to leave out does not send one either. A finding that was already open and has not changed does not send another message; it can send one again if it is closed and then opens again.
  • Credentials. A possible credential is named only "Looks like a credential". The message never shows the secret, its file or its text. Open the finding in Deep scan to see the first and last characters and the length.
  • Wording. A dependency finding "matches a known vulnerability" by version. The message does not say the site is vulnerable. No message does not mean the site is secure.
  • Who gets it. Every member who can open the site gets the message in the bell, in Slack channels and in signed webhooks set up for the organisation, and by email where the service has turned message emails on. Viewers and members see it as well as administrators. Your choice for the email is Security scan findings on the Email settings page, in Critical issues and alerts.
  • The scan did not finish. Organisation administrators get "Security scan of (site) did not finish" (a warning, in the bell and by email only). It says why in words, for example that the scan ran out of time or that its job ended before it stored a result. The findings from the last scan that worked still show. Select Open Deep scan and run the scan again, or contact support if it keeps happening. You get no message when a person cancels a scan, or when the site's repository changed while it ran.
  • No recent scan. Where a site is set to the weekly security scan and no scan of its current repository and branch has finished in 30 days, organisation administrators get "No security scan of (site) in 30 days" (a warning, in the bell and by email only). Its findings may be out of date. The message stays open until a scan of that site finishes, the weekly schedule is turned off, or the site leaves the schedule. It then closes by itself.
  • Not for everyone. The scan did not finish and No recent scan messages go to administrators only. Members and viewers do not get them. Neither goes to Slack or webhooks. If Patching is switched off for your organisation, none of these messages shows, and alerts still waiting to be sent are not sent.

The other four rows:

  • Dependabot alerts (Security): the number of open alerts. Open Dependabot goes to the Dependabot section. The row appears only where Dependabot alerts are available to your organisation.
  • Drupal setup files open to the web (Security, after Dependabot): files such as update.php that a live Drupal site should not serve. It shows Low and Copilot, and it is Included. It appears only where the security scan is on for your organisation, and it opens its own page (see Drupal setup files open to the web).
  • Runtime versions: PHP, Nginx, database (Maintenance): how many components wait for a person. Open runtime settings goes to Configuration, Runtime (see Runtime updates). With no hosting platform connected, the link is Open hosting settings instead.
  • Drupal N upgrade (Upgrades): the site's Upgrade Status readiness for the next major version. See readiness goes to the upgrade readiness panel on the site's Overview tab (see Drupal upgrade readiness). The row appears only where upgrade readiness is available to your organisation and the site's show upgrade readiness switch is on.

A row that cannot run says why, for example that it is turned off for the site, needs a repository (the link is then Choose the repository), or is not available for this Git host or hosting platform. Not checked yet and Partly checked mean Sentinel has no full answer yet. Out of date means the last answer is too old or could not be read again, for example after the Git connection stopped working. Do not read any of them as healthy. If the page cannot load, it says so; try again shortly.