pip-audit
is a tool for scanning Python environments for packages
with known vulnerabilities. It uses the Python Packaging Advisory Database
(https://github.com/pypa/advisory-database) via the
PyPI JSON API as a source
of vulnerability reports.
This project is maintained in part by Trail of Bits with support from Google. This is not an official Google or Trail of Bits product.
- Features
- Installation
- Usage
- Examples
- Troubleshooting
- Tips and Tricks
- Security model
- Licensing
- Contributing
- Code of Conduct
- Support for auditing local environments and requirements-style files
- Support for multiple vulnerability services (PyPI, OSV)
- Support for emitting SBOMs in CycloneDX XML or JSON
- Support for automatically fixing vulnerable dependencies (
--fix
) - Human and machine-readable output formats (columnar, Markdown, JSON)
- Seamlessly reuses your existing local
pip
caches
pip-audit
requires Python 3.7 or newer, and can be installed directly via pip
:
python -m pip install pip-audit
There are multiple third-party packages for pip-audit
. The matrices and badges below
list some of them:
In particular, pip-audit
can be installed via conda
:
conda install -c conda-forge pip-audit
Third-party packages are not directly supported by this project. Please consult your package manager's documentation for more detailed installation guidance.
pip-audit
has an official GitHub Action!
You can install it from the GitHub Marketplace, or add it to your CI manually:
jobs:
pip-audit:
steps:
- uses: pypa/[email protected]
with:
inputs: requirements.txt
See the action documentation for more details and usage examples.
pip-audit
has pre-commit
support.
For example, using pip-audit
via pre-commit
to audit a requirements file:
- repo: https://github.com/pypa/pip-audit
rev: v2.4.9
hooks:
- id: pip-audit
args: ["-r", "requirements.txt"]
ci:
# Leave pip-audit to only run locally and not in CI
# pre-commit.ci does not allow network calls
skip: [pip-audit]
Any pip-audit
arguments documented below can be passed.
You can run pip-audit
as a standalone program, or via python -m
:
pip-audit --help
python -m pip_audit --help
usage: pip-audit [-h] [-V] [-l] [-r REQUIREMENT] [-f FORMAT] [-s SERVICE] [-d]
[-S] [--desc [{on,off,auto}]] [--cache-dir CACHE_DIR]
[--progress-spinner {on,off}] [--timeout TIMEOUT]
[--path PATH] [-v] [--fix] [--require-hashes]
[--index-url INDEX_URL] [--extra-index-url URL]
[--skip-editable] [--no-deps] [-o FILE] [--ignore-vuln ID]
[project_path]
audit the Python environment for dependencies with known vulnerabilities
positional arguments:
project_path audit a local Python project at the given path
(default: None)
optional arguments:
-h, --help show this help message and exit
-V, --version show program's version number and exit
-l, --local show only results for dependencies in the local
environment (default: False)
-r REQUIREMENT, --requirement REQUIREMENT
audit the given requirements file; this option can be
used multiple times (default: None)
-f FORMAT, --format FORMAT
the format to emit audit results in (choices: columns,
json, cyclonedx-json, cyclonedx-xml, markdown)
(default: columns)
-s SERVICE, --vulnerability-service SERVICE
the vulnerability service to audit dependencies
against (choices: osv, pypi) (default: pypi)
-d, --dry-run without `--fix`: collect all dependencies but do not
perform the auditing step; with `--fix`: perform the
auditing step but do not perform any fixes (default:
False)
-S, --strict fail the entire audit if dependency collection fails
on any dependency (default: False)
--desc [{on,off,auto}]
include a description for each vulnerability; `auto`
defaults to `on` for the `json` format. This flag has
no effect on the `cyclonedx-json` or `cyclonedx-xml`
formats. (default: auto)
--cache-dir CACHE_DIR
the directory to use as an HTTP cache for PyPI; uses
the `pip` HTTP cache by default (default: None)
--progress-spinner {on,off}
display a progress spinner (default: on)
--timeout TIMEOUT set the socket timeout (default: 15)
--path PATH restrict to the specified installation path for
auditing packages; this option can be used multiple
times (default: [])
-v, --verbose give more output; this setting overrides the
`PIP_AUDIT_LOGLEVEL` variable and is equivalent to
setting it to `debug` (default: False)
--fix automatically upgrade dependencies with known
vulnerabilities (default: False)
--require-hashes require a hash to check each requirement against, for
repeatable audits; this option is implied when any
package in a requirements file has a `--hash` option.
(default: False)
--index-url INDEX_URL
base URL of the Python Package Index; this should
point to a repository compliant with PEP 503 (the
simple repository API) (default:
https://pypi.org/simple/)
--extra-index-url URL
extra URLs of package indexes to use in addition to
`--index-url`; should follow the same rules as
`--index-url` (default: [])
--skip-editable don't audit packages that are marked as editable
(default: False)
--no-deps don't perform any dependency resolution; requires all
requirements are pinned to an exact version (default:
False)
-o FILE, --output FILE
output results to the given file (default: stdout)
--ignore-vuln ID ignore a specific vulnerability by its vulnerability
ID; this option can be used multiple times (default:
[])
On completion, pip-audit
will exit with a code indicating its status.
The current codes are:
0
: No known vulnerabilities were detected.1
: One or more known vulnerabilities were found.
pip-audit
's exit code cannot be suppressed.
See Suppressing exit codes from pip-audit
for supported alternatives.
pip-audit
supports the --dry-run
flag, which can be used to control whether
an audit (or fix) step is actually performed.
- On its own,
pip-audit --dry-run
skips the auditing step and prints the number of dependencies that would have been audited. - In fix mode,
pip-audit --fix --dry-run
performs the auditing step and prints out the fix behavior (i.e., which dependencies would be upgraded or skipped) that would have been performed.
Audit dependencies for the current Python environment:
$ pip-audit
No known vulnerabilities found
Audit dependencies for a given requirements file:
$ pip-audit -r ./requirements.txt
No known vulnerabilities found
Audit dependencies for a requirements file, excluding system packages:
$ pip-audit -r ./requirements.txt -l
No known vulnerabilities found
Audit dependencies for a local Python project:
$ pip-audit .
No known vulnerabilities found
pip-audit
searches the provided path for various Python "project" files. At the moment, only pyproject.toml
is supported.
Audit dependencies when there are vulnerabilities present:
$ pip-audit
Found 2 known vulnerabilities in 1 package
Name Version ID Fix Versions
---- ------- -------------- ------------
Flask 0.5 PYSEC-2019-179 1.0
Flask 0.5 PYSEC-2018-66 0.12.3
Audit dependencies including descriptions:
$ pip-audit --desc
Found 2 known vulnerabilities in 1 package
Name Version ID Fix Versions Description
---- ------- -------------- ------------ --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Flask 0.5 PYSEC-2019-179 1.0 The Pallets Project Flask before 1.0 is affected by: unexpected memory usage. The impact is: denial of service. The attack vector is: crafted encoded JSON data. The fixed version is: 1. NOTE: this may overlap CVE-2018-1000656.
Flask 0.5 PYSEC-2018-66 0.12.3 The Pallets Project flask version Before 0.12.3 contains a CWE-20: Improper Input Validation vulnerability in flask that can result in Large amount of memory usage possibly leading to denial of service. This attack appear to be exploitable via Attacker provides JSON data in incorrect encoding. This vulnerability appears to have been fixed in 0.12.3. NOTE: this may overlap CVE-2019-1010083.
Audit dependencies in JSON format:
$ pip-audit -f json | python -m json.tool
Found 2 known vulnerabilities in 1 package
[
{
"name": "flask",
"version": "0.5",
"vulns": [
{
"id": "PYSEC-2019-179",
"fix_versions": [
"1.0"
],
"description": "The Pallets Project Flask before 1.0 is affected by: unexpected memory usage. The impact is: denial of service. The attack vector is: crafted encoded JSON data. The fixed version is: 1. NOTE: this may overlap CVE-2018-1000656."
},
{
"id": "PYSEC-2018-66",
"fix_versions": [
"0.12.3"
],
"description": "The Pallets Project flask version Before 0.12.3 contains a CWE-20: Improper Input Validation vulnerability in flask that can result in Large amount of memory usage possibly leading to denial of service. This attack appear to be exploitable via Attacker provides JSON data in incorrect encoding. This vulnerability appears to have been fixed in 0.12.3. NOTE: this may overlap CVE-2019-1010083."
}
]
},
{
"name": "jinja2",
"version": "3.0.2",
"vulns": []
},
{
"name": "pip",
"version": "21.3.1",
"vulns": []
},
{
"name": "setuptools",
"version": "57.4.0",
"vulns": []
},
{
"name": "werkzeug",
"version": "2.0.2",
"vulns": []
},
{
"name": "markupsafe",
"version": "2.0.1",
"vulns": []
}
]
Audit and attempt to automatically upgrade vulnerable dependencies:
$ pip-audit --fix
Found 2 known vulnerabilities in 1 package and fixed 2 vulnerabilities in 1 package
Name Version ID Fix Versions Applied Fix
----- ------- -------------- ------------ ----------------------------------------
flask 0.5 PYSEC-2019-179 1.0 Successfully upgraded flask (0.5 => 1.0)
flask 0.5 PYSEC-2018-66 0.12.3 Successfully upgraded flask (0.5 => 1.0)
Have you resolved a problem with pip-audit
? Help us by contributing to this
section!
In a perfect world, vulnerability feeds would have an infinite signal-to-noise ratio: every vulnerability report would be (1) correct, and (2) applicable to every use of every dependency.
Unfortunately, neither of these is guaranteed: vulnerability feeds are not immune to extraneous or spam reports, and not all uses of a particular dependency map to all potential classes of vulnerabilities.
If your pip-audit
runs produce vulnerability reports that aren't actionable
for your particular application or use case, you can use the --ignore-vuln ID
option to ignore specific vulnerability reports. --ignore-vuln
supports
aliases, so you can use a GHSA-xxx
or CVE-xxx
ID instead of a PYSEC-xxx
ID if the report in question does not have a PYSEC ID.
For example, here is how you might ignore GHSA-w596-4wvx-j9j6, which is a
common source of noisy vulnerability reports and false positives for users of
pytest
:
# Run the audit as normal, but exclude any reports that match GHSA-w596-4wvx-j9j6
$ pip-audit --ignore-vuln GHSA-w596-4wvx-j9j6
The --ignore-vuln ID
option works with all other dependency resolution
and auditing options, meaning that it should function correctly with
requirements-style inputs, alternative vulnerability feeds, and so forth.
It can also be passed multiple times, to ignore multiple reports:
# Run the audit as normal, but exclude any reports that match these IDs
$ pip-audit --ignore-vuln CVE-XXX-YYYY --ignore-vuln CVE-ZZZ-AAAA
Depending on how you're using it, pip-audit
may have to perform its
own dependency resolution, which can take roughly as long as pip install
does for a project. See the security model for an explanation.
You have two options for avoiding dependency resolution: audit a pre-installed environment, or ensure that your dependencies are already fully resolved.
If you know that you've already fully configured an environment equivalent
to the one that pip-audit -r requirements.txt
would audit, you can simply
reuse it:
# Note the absence of any "input" arguments, indicating that the environment is used.
$ pip-audit
# Optionally filter out non-local packages, for virtual environments:
$ pip-audit --local
Alternatively, if your input is fully pinned (and optionally hashed), you
can tell pip-audit
to skip dependency resolution with either --no-deps
(pinned without hashes) or --require-hashes
(pinned including hashes).
The latter is equivalent to pip
's
hash-checking mode
and is preferred, since it offers additional integrity.
# fails if any dependency is not fully pinned
$ pip-audit --no-deps -r requirements.txt
# fails if any dependency is not fully pinned *or* is missing hashes
$ pip-audit --require-hashes -r requirements.txt
pipenv
uses both a Pipfile
and Pipfile.lock
file to track and freeze dependencies
instead of a requirements.txt
file. pip-audit
cannot process the Pipfile[.lock]
files directly, however, these can be converted to a supported requirements.txt
file
that pip-audit
can run against. Use a Python tool, such as
pipfile-requirements
, to
convert your Pipfile[.lock]
to a requirements.txt
file and then run
pip-audit
against the generated requirements file.
$ pipenv install --dev pipfile-requirements
$ pipenv run pipfile2req Pipfile.lock > requirements.txt
$ pipenv run pip-audit -r requirements.txt
The requirements.txt
file needs to be kept up to date with pipenv
when running
pip-audit
locally like this.
pip-audit
intentionally does not support internally suppressing its own
exit codes.
Users who need to suppress a failing pip-audit
invocation can use
one of the standard shell idioms for doing so:
pip-audit || true
or, to exit entirely:
pip-audit || exit 0
The exit code can also be captured and handled explicitly:
pip-audit
exitcode="${?}"
# do something with ${exitcode}
See Exit codes for a list of potential codes that need handling.
This section exists to describe the security assumptions you can and must not
make when using pip-audit
.
TL;DR: If you wouldn't pip install
it, you should not pip audit
it.
pip-audit
is a tool for auditing Python environments for packages with
known vulnerabilities. A "known vulnerability" is a publicly reported flaw in
a package that, if uncorrected, might allow a malicious actor to perform
unintended actions.
pip-audit
can protect you against known vulnerabilities by telling
you when you have them, and how you should upgrade them. For example,
if you have somepackage==1.2.3
in your environment, pip-audit
can tell
you that it needs to be upgraded to 1.2.4
.
You can assume that pip-audit
will make a best effort to fully resolve
all of your Python dependencies and either fully audit each or explicitly
state which ones it has skipped, as well as why it has skipped them.
pip-audit
is not a static code analyzer. It analyzes dependency trees,
not code, and it cannot guarantee that arbitrary dependency resolutions
occur statically. To understand why this is, refer to Dustin Ingram's
excellent post on dependency resolution in Python.
As such: you must not assume that pip-audit
will defend you against
malicious packages. In particular, it is incorrect to treat
pip-audit -r INPUT
as a "more secure" variant of pip-audit
. For all intents
and purposes, pip-audit -r INPUT
is functionally equivalent to
pip install -r INPUT
, with a small amount of non-security isolation to
avoid conflicts with any of your local environments.
pip-audit
is first and foremost a auditing tool for Python packages.
You must not assume that pip-audit
will detect or flag "transitive"
vulnerabilities that might be exposed through Python packages, but are not
actually part of the package itself. For example, pip-audit
's vulnerability
information sources are unlikely to include an advisory for a vulnerable shared
library that a popular Python package might use, since the Python package's
version is not strongly connected to the shared library's version.
pip-audit
is licensed under the Apache 2.0 License.
pip-audit
reuses and modifies examples from
resolvelib
, which is licensed under
the ISC license.
See the contributing docs for details.
Everyone interacting with this project is expected to follow the PSF Code of Conduct.