ex_check_ngis a community-maintained fork ofex_check(dormant since 2024). Module namespace (ExCheck) and themix checktask are unchanged — drop-in replacement. Install as{:ex_check_ng, "~> 1.0", only: [:dev], runtime: false}.
Run all code checking tools with a single convenient mix check command.
Takes seconds to setup, saves hours in the long term.
- Comes out of the box with a predefined set of curated tools
- Delivers results faster by running tools in parallel and catching all issues in one go
- Checks the project consistently on every developer's local machine & on the CI
- Runs only the tools & tests that have failed in the last run
- Fixes issues automatically in the fix mode
Sports powerful features to enable ultimate flexibility.
- Add custom mix tasks, shell scripts and commands via configuration file
- Enhance you CI workflow to report status, retry random failures or autofix issues
- Empower umbrella projects with parallel recursion over child apps
- Design complex parallel workflows with cross-tool deps
Takes care of the little details, so you don't have to.
- Compiles the project and collects compilation warnings in one go
- Ensures that output from tools is ANSI formatted & colorized
- Retries ExUnit with the
--failedflag
Read more in the introductory "One task to rule all Elixir analysis & testing tools" article.
Getting started
Add ex_check_ng dependency in mix.exs:
def deps do
[
{:ex_check_ng, "~> 1.0", only: [:dev], runtime: false}
]
end
Fetch the dependency:
mix deps.get
Run the check:
mix check
That's it - mix check will detect and run all the available tools.
Community tools
If you want to take advantage of community curated tools, add following dependencies in mix.exs:
def deps do
[
{:credo, ">= 0.0.0", only: [:dev], runtime: false},
{:dialyxir, ">= 0.0.0", only: [:dev], runtime: false},
{:doctor, ">= 0.0.0", only: [:dev], runtime: false},
{:ex_doc, ">= 0.0.0", only: [:dev], runtime: false},
{:gettext, ">= 0.0.0", only: [:dev], runtime: false},
{:sobelow, ">= 0.0.0", only: [:dev], runtime: false},
{:mix_audit, ">= 0.0.0", only: [:dev], runtime: false}
]
end
You may also generate .check.exs to adjust the check:
mix check.gen.config
Among others, this allows to permanently disable specific tools and avoid the skipped notices:
[
tools: [
{:dialyzer, false},
{:sobelow, false}
]
]
Usage rules for coding agents
ex_check_ng ships usage rules — concise, authoritative
guidance written for LLM coding agents (Claude Code, Cursor, ...) on how to drive mix check,
including using mix check --format agent for machine-readable output.
It is recommended to sync these into your project's agent rules file so agents automatically pick
the right flags. Add usage_rules:
def deps do
[
{:usage_rules, "~> 1.2", only: [:dev], runtime: false}
]
end
Configure the sync in mix.exs (the config is the source of truth):
def project do
[
# ...
usage_rules: [
file: "AGENTS.md",
usage_rules: [:ex_check_ng]
]
]
end
Then run:
mix deps.get
mix usage_rules.sync
This keeps an ex_check_ng section in your AGENTS.md in sync with the rules shipped by the
package, so coding agents run mix check --format agent instead of parsing human-oriented output.
Output formats
mix check renders its results through pluggable reporters selected with --format (default
pretty). Batched formats (agent, json, junit) may be written to a file with --output PATH
instead of stdout.
| Format | Use case |
|---|---|
pretty | Live colored terminal output for humans (default). |
agent | JSON status header + raw failure blocks, optimized for LLM/agent consumption. |
json | Single machine-readable JSON object describing the whole run. |
github | GitHub Actions log groups + a Markdown summary appended to $GITHUB_STEP_SUMMARY. |
junit | JUnit XML report for CI systems like GitLab CI and Jenkins. |
mix check --format agent # LLM-friendly
mix check --format json --output check.json # machine-readable file
mix check --format github # inside a GitHub Actions job
mix check --format junit --output report.xml # JUnit report for CI
github and pretty stream to the terminal and reject --output. See
Continuous Integration for wiring the CI formats into a pipeline.
Local-only fix mode
You should keep local and CI configuration as consistent as possible by putting together the project-specific .check.exs. Still, you may introduce local-only config by creating the ~/.check.exs file. This may be useful to enforce global flags on all local runs. For example, the following config will enable the fix mode in local (writable) environment:
[
fix: true
]
You may also enable the fix mode on the CI.
Documentation
Learn more about the tools included in the check as well as its workflow, configuration and options on HexDocs or by running mix help check.
Want to write your own code check? Get yourself started by reading the "Writing your first Elixir code check" article.
Continuous Integration
On GitHub Actions, --format github adds collapsible
per-tool log groups and a run summary on the job page:
# .github/workflows/check.yml
name: check
on: [push, pull_request]
jobs:
check:
runs-on: ubuntu-24.04
steps:
- uses: actions/checkout@v4
- uses: erlef/setup-beam@v1
with:
elixir-version: "1.20"
otp-version: "28"
- run: mix deps.get
- run: mix check --format github
For CI systems with a JUnit test report (GitLab CI, Jenkins, ...), use mix check --format junit --output report.xml. See Output formats for the full list.
ex_check uses itself on the CI — see our own workflow.
Autofixing
You may automatically fix and commit back trivial issues by triggering the fix mode on the CI as well. In order to do so, you'll need a CI script or workflow similar to the example below:
mix check --fix && \
git diff-index --quiet HEAD -- && \
git config --global user.name 'Autofix' && \
git config --global user.email 'autofix@example.com' && \
git add --all && \
git commit --message "Autofix" && \
git push
First, we perform the check in the fix mode. Then, if no unfixable issues have occurred and if fixes were actually made, we proceed to commit and push these fixes.
Of course your CI will need to have write permissions to the source repository.
Random failures
You may take advantage of the automatic retry feature to efficiently re-run failed tools & tests multiple times. For instance, following shell command runs check up to three times: mix check || mix check || mix check. And here goes an alternative without the logical operators:
mix check
mix check --retry
mix check --retry
This will work as expected because the --retry flag will ensure that only failed tools are executed, resulting in no-op if previous run has succeeded.
Troubleshooting
Duplicate builds
If, as suggested above, you've added ex_check and curated tools to only: [:dev], you're keeping the test environment reserved for ex_unit. While a clean setup, it comes at the expense of Mix having to compile your app twice - in order to prepare :test build just for ex_unit and :dev build for other tools. This costs precious time both on local machine and on the CI. It may also cause issues if you set MIX_ENV=test, which is a common practice on the CI.
You may avoid this issue by running mix check and all the tools it depends on in the test environment. In such case you may want to have the following config in mix.exs:
def cli do
[
preferred_envs: [
check: :test,
credo: :test,
dialyzer: :test,
doctor: :test,
docs: :test,
format: :test,
sobelow: :test,
"deps.audit": :test
]
]
end
def deps do
[
{:credo, ">= 0.0.0", only: [:test], runtime: false},
{:dialyxir, ">= 0.0.0", only: [:test], runtime: false},
{:doctor, ">= 0.0.0", only: [:test], runtime: false},
{:ex_check_ng, "~> 1.0", only: [:test], runtime: false},
{:ex_doc, ">= 0.0.0", only: [:dev, :test], runtime: false},
{:mix_audit, ">= 0.0.0", only: [:test], runtime: false}
{:sobelow, ">= 0.0.0", only: [:test], runtime: false},
]
end
And the following in .check.exs:
[
tools: [
{:compiler, env: %{"MIX_ENV" => "test"}},
{:ex_doc, env: %{"MIX_ENV" => "test"}}
{:formatter, env: %{"MIX_ENV" => "test"}},
]
]
Above setup will consistently check the project using just the test build, both locally and on the CI.
unused_deps false negatives
You may encounter an issue with the unused_deps check failing on the CI while passing locally, caused by fetching only dependencies for specific env. If that happens, remove the --only test (or similar) from your mix deps.get invocation on the CI to fix the issue.
Changelog
See CHANGELOG.md.
Copyright and License
Copyright (c) 2019 Karol Słuszniak
This work is free. You can redistribute it and/or modify it under the terms of the MIT License. See the LICENSE.md file for more details.