Dependency Selector Syntax & Querying

Dependency Selector Syntax & Querying

Select CLI Version:


The npm query command exposes a new dependency selector syntax (informed by & respecting many aspects of the CSS Selectors 4 Spec) which:

  • Standardizes the shape of, & querying of, dependency graphs with a robust object model, metadata & selector syntax
  • Leverages existing, known language syntax & operators from CSS to make disparate package information broadly accessible
  • Unlocks the ability to answer complex, multi-faceted questions about dependencies, their relationships & associative metadata
  • Consolidates redundant logic of similar query commands in npm (ex. npm fund, npm ls, npm outdated, npm audit ...)

Dependency Selector Syntax v1.0.0


  • there is no "type" or "tag" selectors (ex. div, h1, a) as a dependency/target is the only type of Node that can be queried
  • the term "dependencies" is in reference to any Node found in a tree returned by Arborist


  • > direct descendant/child
  • any descendant/child
  • ~ sibling


  • * universal selector
  • #<name> dependency selector (equivalent to [name="..."])
  • #<name>@<version> (equivalent to [name=<name>]:semver(<version>))
  • , selector list delimiter
  • . dependency type selector
  • : pseudo selector

Dependency Type Selectors

  • .prod dependency found in the dependencies section of package.json, or is a child of said dependency
  • .dev dependency found in the devDependencies section of package.json, or is a child of said dependency
  • .optional dependency found in the optionalDependencies section of package.json, or has "optional": true set in its entry in the peerDependenciesMeta section of package.json, or a child of said dependency
  • .peer dependency found in the peerDependencies section of package.json
  • .workspace dependency found in the workspaces section of package.json
  • .bundled dependency found in the bundleDependencies section of package.json, or is a child of said dependency

Pseudo Selectors

  • :not(<selector>)
  • :has(<selector>)
  • :is(<selector list>)
  • :root matches the root node/dependency
  • :scope matches node/dependency it was queried against
  • :empty when a dependency has no dependencies
  • :private when a dependency is private
  • :link when a dependency is linked (for instance, workspaces or packages manually linked
  • :deduped when a dependency has been deduped (note that this does not always mean the dependency has been hoisted to the root of node_modules)
  • :overridden when a dependency has been overridden
  • :extraneous when a dependency exists but is not defined as a dependency of any node
  • :invalid when a dependency version is out of its ancestors specified range
  • :missing when a dependency is not found on disk
  • :semver(<spec>, [selector], [function]) match a valid node-semver version or range to a selector
  • :path(<path>) glob matching based on dependencies path relative to the project
  • :type(<type>) based on currently recognized types
  • :outdated(<type>) when a dependency is outdated
:semver(<spec>, [selector], [function])

The :semver() pseudo selector allows comparing fields from each node's package.json using semver methods. It accepts up to 3 parameters, all but the first of which are optional.

  • spec a semver version or range
  • selector an attribute selector for each node (default [version])
  • function a semver method to apply, one of: satisfies, intersects, subset, gt, gte, gtr, lt, lte, ltr, eq, neq or the special function infer (default infer)

When the special infer function is used the spec and the actual value from the node are compared. If both are versions, according to semver.valid(), eq is used. If both values are ranges, according to !semver.valid(), intersects is used. If the values are mixed types satisfies is used.

Some examples:

  • :semver(^1.0.0) returns every node that has a version satisfied by the provided range ^1.0.0
  • :semver(16.0.0, :attr(engines, [node])) returns every node which has an engines.node property satisfying the version 16.0.0
  • :semver(1.0.0, [version], lt) every node with a version less than 1.0.0

The :outdated pseudo selector retrieves data from the registry and returns information about which of your dependencies are outdated. The type parameter may be one of the following:

  • any (default) a version exists that is greater than the current one
  • in-range a version exists that is greater than the current one, and satisfies at least one if its dependents
  • out-of-range a version exists that is greater than the current one, does not satisfy at least one of its dependents
  • major a version exists that is a semver major greater than the current one
  • minor a version exists that is a semver minor greater than the current one
  • patch a version exists that is a semver patch greater than the current one

In addition to the filtering performed by the pseudo selector, some extra data is added to the resulting objects. The following data can be found under the queryContext property of each node.

  • versions an array of every available version of the given node
  • outdated.inRange an array of objects, each with a from and versions, where from is the on-disk location of the node that depends on the current node and versions is an array of all available versions that satisfies that dependency. This is only populated if :outdated(in-range) is used.
  • outdated.outOfRange an array of objects, identical in shape to inRange, but where the versions array is every available version that does not satisfy the dependency. This is only populated if :outdated(out-of-range) is used.

Some examples:

  • :root > :outdated(major) returns every direct dependency that has a new semver major release
  • .prod:outdated(in-range) returns production dependencies that have a new release that satisfies at least one of its edges in

Attribute Selectors

The attribute selector evaluates the key/value pairs in package.json if they are Strings.

  • [] attribute selector (ie. existence of attribute)
  • [attribute=value] attribute value is equivalant...
  • [attribute~=value] attribute value contains word...
  • [attribute*=value] attribute value contains string...
  • [attribute|=value] attribute value is equal to or starts with...
  • [attribute^=value] attribute value starts with...
  • [attribute$=value] attribute value ends with...

Array & Object Attribute Selectors

The generic :attr() pseudo selector standardizes a pattern which can be used for attribute selection of Objects, Arrays or Arrays of Objects accessible via Arborist's Node.package metadata. This allows for iterative attribute selection beyond top-level String evaluation. The last argument passed to :attr() must be an attribute selector or a nested :attr(). See examples below:


/* return dependencies that have a `scripts.test` containing `"tap"` */
*: attr(scripts, [test~=tap]);

Nested Objects

Nested objects are expressed as sequential arguments to :attr().

/* return dependencies that have a testling config for opera browsers */
*: attr(testling, browsers, [~=opera]);


Arrays specifically uses a special/reserved . character in place of a typical attribute name. Arrays also support exact value matching when a String is passed to the selector.

Example of an Array Attribute Selection:
/* removes the distinction between properties & arrays */
/* ie. we'd have to check the property & iterate to match selection */
*:attr(contributors, :attr([name~=Jordan]))
Example of an Array matching directly to a value:
/* return dependencies that have the exact keyword "react" */
/* this is equivalent to `*:keywords([value="react"])` */
*: attr([keywords=react]);
Example of an Array of Objects:
/* returns */
*: attr(contributors, [email=ruyadorno]);


Dependency groups are defined by the package relationships to their ancestors (ie. the dependency types that are defined in package.json). This approach is user-centric as the ecosystem has been taught to think about dependencies in these groups first-and-foremost. Dependencies are allowed to be included in multiple groups (ex. a prod dependency may also be a dev dependency (in that it's also required by another dev dependency) & may also be bundled - a selector for that type of dependency would look like: *

  • .prod
  • .dev
  • .optional
  • .peer
  • .bundled
  • .workspace

Please note that currently workspace deps are always prod dependencies. Additionally the .root dependency is also considered a prod dependency.

Programmatic Usage

  • Arborist's Node Class has a .querySelectorAll() method
    • this method will return a filtered, flattened dependency Arborist Node list based on a valid query selector
const Arborist = require("@npmcli/arborist");
const arb = new Arborist({});
// root-level
arb.loadActual().then(async (tree) => {
// query all production dependencies
const results = await tree.querySelectorAll(".prod");
// iterative
arb.loadActual().then(async (tree) => {
// query for the deduped version of react
const results = await tree.querySelectorAll("#react:not(:deduped)");
// query the deduped react for git deps
const deps = await results[0].querySelectorAll(":type(git)");

See Also