Comments on each pull request with what it does to the tests, without running them.
The action scans the pull request's base and head with test-inventory, which reads Playwright, Vitest, Jest, Mocha, Jasmine, Cypress, WebdriverIO, node:test and Bun tests from source. It needs no build, no browsers and no install of your project, and it never fails a check.
The comment shows what a reviewer should look at:
- Won't run: a new test file no runner config collects, or tests a config change stopped collecting.
- Turned off: existing tests that now skip, grouped by the condition or reason, and tests that skipped only on a condition, such as a missing secret, and now always skip.
- Removed: tests taken out of a file that gained none.
- Test files deleted.
- No assertions left: tests that had
expectcalls and now have none.
The rest is collapsed:
- tests rewritten or replaced in the same file;
- renamed or moved tests;
- tests whose helpers or source changed;
- fewer assertions;
- added tests.
When nothing needs a look, the action posts no comment, or updates its earlier one to say so.
A results reporter says which tests passed and failed in a run. This action reads the tests themselves, before anything runs, so it covers what a run can't show:
- tests no runner config collects, which no run reports;
- the condition and reason that turn a test off;
- a renamed test as a rename, rather than a removal and an addition;
- a test that lost its assertions, which a run still counts as passed;
- pull requests whose tests didn't run at all, such as when a path filter skipped the job.
It needs no reporter setup, results file or account, and it runs none of the pull request's code, so it works the same on pull requests from forks. Keep your reporter for failures and timings: the two answer different questions.
Two workflows: one scans with read-only access, and one posts the comment once the scan has finished. The second one never checks out the pull request's code, so the comment works for pull requests from forks too.
# .github/workflows/test-changes.yml
name: Test changes
on: pull_request
permissions:
contents: read
jobs:
scan:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
with:
persist-credentials: false
- id: tests
uses: Huldoser/test-inventory-action@1a971576a927060370e6a74ab307d0bf15b3eba8 # v0.2.3
- uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
with:
name: test-inventory-report
path: ${{ steps.tests.outputs.report-dir }}
if-no-files-found: ignore# .github/workflows/test-changes-comment.yml
name: Test changes comment
on:
workflow_run: # zizmor: ignore[dangerous-triggers]
workflows: [Test changes]
types: [completed]
permissions:
actions: read
pull-requests: write
jobs:
comment:
if: github.event.workflow_run.event == 'pull_request' && github.event.workflow_run.conclusion == 'success'
runs-on: ubuntu-latest
steps:
- uses: Huldoser/test-inventory-action@1a971576a927060370e6a74ab307d0bf15b3eba8 # v0.2.3
with:
mode: commentPin the action to a full commit SHA, as in the examples, and let Dependabot or your workflow lockfile move the pin.
The scan measures the pull request against its base branch as it is now: actions/checkout checks out the pull
request merged into that branch, and the action compares the merge with the branch. A workflow that checks out the
pull request's head commit instead gets it compared with where it last met the base branch, which the action asks
GitHub for with the job's token.
The comment workflow reads the report the scan uploaded as data and checks it field by field. It posts only if the pull request is still at the commit that was scanned, and only edits a comment a bot wrote.
zizmor's dangerous-triggers audit flags every workflow_run workflow. This one checks out no code, reads the report
as data and posts only for the commit that was scanned, so the example ignores that finding on its trigger line.
A pull request from a fork runs the fork's own copy of the scan workflow, so its author controls what the report says and can drop a step that gates on the outputs. The comment workflow keeps the report from posting links, mentions or markup of its own, but it can't tell an honest report from a made-up one. On a fork's pull request, read the comment as a guide to the diff, not as proof, and keep any gate in a workflow the fork can't change. GitHub asks a maintainer to approve workflow runs from first-time contributors.
| Input | Default | What it does |
|---|---|---|
mode |
scan |
scan in the pull_request workflow, comment in the workflow_run workflow. |
github-token |
${{ github.token }} |
Downloads the report and writes the comment, in comment mode. In scan mode, fetches the base commit and finds where a checked-out head met the base branch. |
artifact-name |
test-inventory-report |
The artifact comment mode downloads; match the scan's upload. |
timeout-minutes |
10 |
The time the scan may take; past it the action stops and writes no report. |
wont-run, turned-off, removed and asserts-nothing count the tests in those sections, and report-dir is the
folder with report.json. The action never fails because of what it finds. To make a finding block a pull request,
add a step of your own:
- if: steps.tests.outputs.wont-run != '0'
run: |
echo "::error::Some tests in this pull request won't run; see the job summary."
exit 1The scan reads source and configs; it runs nothing. These are out of its reach:
- Tests registered at run time, configs computed from environment variables, and settings it can't work out, which it reports as unknown rather than guess.
- A config named only in a CI workflow, or chosen by a wrapper script, rather than in a package.json script.
- A runner that no package.json lists and that has no config file of its usual name, other than Node's and Bun's built-in runners.
- Very large repositories may pass the time limit; the job summary then says so.
Known limits has the scanner's full list.
Node.js 22.12 or newer on the runner, which every GitHub-hosted image has. Self-hosted runners may need
actions/setup-node first. The action needs network access to npm to install test-inventory.
MIT, copyright © 2026 Andrey Rychkov.