Add a new improver

This tutorial contains all the things one should know to quickly implement an improver. A lot of internal sausage about improvers could be found inside the vulnerabilites/ file. Make sure to go through Improver Overview before you begin writing one.


  1. Locate the importer that this improver will be improving data of at vulnerabilities/importers/ file.

  2. Create a new improver subclass inheriting from the Improver superclass defined in vulnerabilites.improver. It is conventional to end an improver name with Improver.

  3. Implement the interesting_advisories property to return a QuerySet of imported data (Advisory) you are interested in.

  4. Implement the get_inferences method to return an iterable of Inference objects for the given AdvisoryData

  5. Add the newly created improver to the improvers registry at vulnerabilites/improvers/


Before writing an improver, it is important to familiarize yourself with the following concepts.


Importers are responsible for scraping vulnerability data from various data sources without creating a complete relational model between vulnerabilites, their fixes and store them in a structured fashion. These data are stored in the Advisory model and can be converted to an equivalent AdvisoryData for various use cases. See Importer Overview for a brief overview on importers.

Importer Prerequisites

Improvers consume data produced by importers, thus it is important to familiarize yourself with Importer Prerequisites


Inferences express the contract between the improvers and the improve runner framework. An inference is supposed to contain data points about a vulnerability without any uncertainties, which means, one inference will target one vulnerability with the specific relevant affected and fixed packages (in the form of PackageURLs) There is no notion of version ranges here, all package versions must be explicitly specified.

Because this concrete relationship is hardly available anywhere on the upstream, we have to infer these values, thus the name. As infering something is not always perfect, an Inference also comes with a confidence score.


All the Improvers must inherit from Improver superclass and implement the interesting_advisories property and the get_inferences method.

Writing an improver

Locate the Source File

If the improver will be working on data imported by an specific importer, it will sit in the same file at vulnerabilites/importers/ Otherwise, if it is a generic improver, create a new file vulnerabilites/improvers/

Explore Package Managers (Optional)

If your Improver depends on the discrete versions of a package, the package managers’ VersionAPI located at vulnerabilites/ could come in handy. You’ll need to instantiate the relevant VersionAPI in the improver’s constructor and use them later in the implemented methods. See an already implemented improver (NginxBasicImprover) for an example usage.

Implement the interesting_advisories Property

This property is supposed to return a QuerySet of Advisory on which the Improver is interested to work on.

For example, if the improver is interested to work on Advisories imported by ExampleImporter, the property can be implemented as

class ExampleBasicImprover(Improver):

    def interesting_advisories(self) -> QuerySet:
        return Advisory.objects.filter(created_by=ExampleImporter.qualified_name)

Implement the get_inferences Method

The framework calls get_inferences method for every AdvisoryData that is obtained from the Advisory QuerySet returned by the interesting_advisories property.

It is expected to return an iterable of Inference objects for the given AdvisoryData. To avoid storing a lot of Inferences in memory, it is nicer to yield from this method.

A very simple Improver that processes all Advisories to create the minimal relationships that can be obtained by existing data can be found at vulnerabilites/improvers/ It is an example of a generic improver, for more sophisticated and targetted one, you can look at an already implemented improver (for eg, in vulnerabilites/importers/

Improvers are not limited to improving discrete versions, they may also improve aliases. One such example, improving the importer written in the importer tutorial, is shown below.

from datetime import datetime
from datetime import timezone
from typing import Iterable

import requests
from django.db.models.query import QuerySet
from packageurl import PackageURL
from univers.version_range import NginxVersionRange
from univers.versions import SemverVersion

from vulnerabilities.importer import AdvisoryData
from vulnerabilities.improver import MAX_CONFIDENCE
from vulnerabilities.improver import Improver
from vulnerabilities.improver import Inference
from vulnerabilities.models import Advisory
from vulnerabilities.severity_systems import SCORING_SYSTEMS

class ExampleImporter(Importer):

class ExampleAliasImprover(Improver):
    def interesting_advisories(self) -> QuerySet:
        return Advisory.objects.filter(created_by=ExampleImporter.qualified_name)

    def get_inferences(self, advisory_data) -> Iterable[Inference]:
        for alias in advisory_data.aliases:
            new_aliases = fetch_additional_aliases(alias)
            aliases = new_aliases + [alias]
            yield Inference(aliases=aliases, confidence=MAX_CONFIDENCE)

def fetch_additional_aliases(alias):
    alias_map = {
        "CVE-2021-23017": ["PYSEC-1337", "CERTIN-1337"],
        "CVE-2021-1234": ["ANONSEC-1337", "CERTDES-1337"],
    return alias_map.get(alias)


Use make valid to format your new code using black and isort automatically.
Use make check to check for formatting errrors.

Register the Improver

Finally, register your improver in the improver registry at vulnerabilites/improvers/

 from vulnerabilities import importers
 from vulnerabilities.improvers import default


 IMPROVERS_REGISTRY = {x.qualified_name: x for x in IMPROVERS_REGISTRY}

Congratulations! You’ve written your first improver.

Run Your First Improver

If everything went fine, you’ll see your improver in the list of available improvers

 $ ./ improve --list

 Vulnerability data can be processed by these available improvers:

Before running the improver, make sure you have imported the data. An improver cannot improve if there is nothing imported.

$ ./ import vulnerabilities.importers.example.ExampleImporter

Importing data using vulnerabilities.importers.example.ExampleImporter
Successfully imported data using vulnerabilities.importers.example.ExampleImporter

Now, run the improver

$ ./ improve vulnerabilities.importers.example.ExampleAliasImprover

 Improving data using vulnerabilities.importers.example.ExampleAliasImprover
 Successfully improved data using vulnerabilities.importers.example.ExampleAliasImprover

See Command Line Interface for command line usage instructions.

Enable Debug Logging (Optional)

For more visibility, turn on debug logs in vulnerablecode/

DEBUG = True
    'version': 1,
    'disable_existing_loggers': False,
    'handlers': {
        'console': {
            'class': 'logging.StreamHandler',
    'root': {
        'handlers': ['console'],
        'level': 'DEBUG',

Invoke the improve command now and you’ll see (in a fresh database, after importing)

$ ./ improve vulnerabilities.importers.example.ExampleAliasImprover

Improving data using vulnerabilities.importers.example.ExampleAliasImprover
Running improver: vulnerabilities.importers.example.ExampleAliasImprover
Improving advisory id: 1
New alias for <Vulnerability: VULCOID-23dd9060-3bc0-4454-bfbd-d16c08a966a6>: PYSEC-1337
New alias for <Vulnerability: VULCOID-23dd9060-3bc0-4454-bfbd-d16c08a966a6>: CVE-2021-23017
New alias for <Vulnerability: VULCOID-23dd9060-3bc0-4454-bfbd-d16c08a966a6>: CERTIN-1337
Improving advisory id: 2
New alias for <Vulnerability: VULCOID-fae4e06e-4815-45fe-ae95-8d2356ffb5b9>: CERTDES-1337
New alias for <Vulnerability: VULCOID-fae4e06e-4815-45fe-ae95-8d2356ffb5b9>: ANONSEC-1337
New alias for <Vulnerability: VULCOID-fae4e06e-4815-45fe-ae95-8d2356ffb5b9>: CVE-2021-1234
Finished improving using vulnerabilities.importers.example.ExampleAliasImprover.
Successfully improved data using vulnerabilities.importers.example.ExampleAliasImprover


Even though CVE-2021-23017 and CVE-2021-1234 are not supplied by this improver yet it shows them because we left out running the DefaultImprover in the example. The DefaultImprover inserts minimal data found via the importers in the database (Here, the above two CVEs). Run importer, DefaultImprover and then your improver in this sequence to avoid this anomaly.