← All posts

Finding CVEs is easy. Knowing which to fix is not.

Every team wants to fix its CVEs, but almost no scan comes back empty. A count of open CVEs does not tell you what to do next. Three questions do: does this CVE affect you, can you fix it today, and which fix is due first?

Zero CVEs is rare

Container images with a high or critical CVE

87%

Nearly every image you ship carries one.

Sysdig

Behind the latest major version

215 days

The median dependency, and one in two services uses an unmaintained library.

Datadog

Take an example storefront. Its CI scans the container image on every build, and today the scanner reports 214 CVEs. The team wants to fix them all, but it cannot fix 214 things this week. It has to choose, and the count does not help it choose.

Does it affect you?

Some CVEs are in a package you ship but in code you never run, or in a feature you do not use. Datadog found that just 18% of critical CVEs stay critical once you add context about how the code runs. When a CVE does not affect you, record that, with a reason, so nobody checks it again next month.

Can you fix it today?

For most CVEs in a container image, nobody has released a fix yet. Those stay on the list, but you cannot act on them today. The work you can do this week is the short list of CVEs that already have a fixed version:

The storefront's 214 CVEs

Found by the scan214
Affect the storefront191
Have a fix today9
Due first1

23 CVEs are marked as not affecting the storefront. 182 have no fix yet.

Which fix is due first?

Severity alone does not order the list. A high CVE you found a month ago can be due before a critical one you found this morning. So each fixable CVE needs a deadline, counted from the day you first saw it. A common policy is 7 days for critical and 30 days for high.

The 9 CVEs you can fix Deadline
Critical, in the JWT library
Found 12 days ago
5 days overdue
High, in the image library
Found 26 days ago
Due in 4 days
7 more, all high Within deadline

Deadlines also tell you if you keep up. Datadog measured a median time to fix library CVEs of 62 days for Maven, 46 for NuGet and 19 for npm. If your average is longer than your deadlines, you will fall behind, whatever the scan count says.

How Drape does it

Drape reads your scanner's results in CI and answers the three questions for every CVE:

  • Does it affect you? You can mark a CVE as not affected, not applicable or a false positive, with a reason. Drape keeps the history and stops its deadline.
  • Can you fix it today? Drape checks each CVE for a fixed version, and sorts it into "to fix" or "no fix". A CVE that the vendor has decided never to patch, or marks as not affecting you, never fails a pull request.
  • Which is due first? Each fixable CVE gets a deadline, 7 days for critical and 30 for high until you set your own. A pull request fails when a fixable CVE is past its deadline. Drape also shows your average days to fix over the last 30 days.

Most teams silence a CVE with a line in an ignore file, committed next to the code. That line ignores the CVE forever. Nobody records why, nobody is asked to look again, and the CVE drops out of every report. A suppression in Drape is a decision you can check later:

Ignore file in the repository Suppression in Drape
Records who and why No Yes
With a reason, a note and a ticket link
Asks you to look again No Yes
On a review interval you choose
Keeps the CVE visible No Yes
Listed as suppressed, with its history
Ends on a date and tells you No
Even where a line can expire, the CVE just reappears in the scan
Yes
The CVE reopens, and a paused deadline picks up where it stopped

It applies to one repository, and you can narrow it to a package or an image. The reason decides what happens to the deadline:

Suppressed CVE Reason Deadline
Critical, XML parser
"The storefront never parses XML." Review every 90 days.
Not applicable Stopped
High, PDF library
"Upgrade with the invoice rewrite." Linked to a ticket.
Technical debt Still running

Accepting a risk does not hide it. When you choose to leave a fixable CVE for later, as technical debt or as your own "won't fix", its deadline keeps running, and Drape flags a suppression whose review is overdue. Every suppression and its removal shows up in the repository's activity feed.

So the scan of the storefront does not end with "214 CVEs". It ends with one thing to do:

  1. CI uploads the scan and Drape sorts it

    $ drape upload scan grype.json
    Scan Summary (grype.json)
      Vulnerabilities: 214
      Repository:      storefront (main): 9 to fix (1 past SLA),
                       182 with no fix, 23 suppressed
  2. Your agent asks what is due through Drape's MCP server

    > Which CVE should we fix first in storefront?
    ● drape · list_vulnerabilities(repo="storefront", sla="overdue")
      critical, JWT library: fixed version available, 5 days past its deadline
    ● Upgrading the JWT library to the fixed version and running the tests.
  3. The fix merges and the overdue CVE is gone

    $ drape upload scan grype.json
      Repository:      storefront (main): 8 to fix,
                       182 with no fix, 23 suppressed

You will not get to zero CVEs. You can make sure the next fix you make is the right one.

Know which CVE to fix first.

14-day free trial. No credit card required.