Use
Pull request checks (GitHub Action)
Scan only what a pull request changes with the 0 Action, and show the findings in GitHub's Security tab.
The 0 Action scans a pull request and uploads the result to GitHub code scanning, so findings appear in the repository’s Security tab and as annotations on the pull request. It needs no server: it installs a verified release of 0, runs 0 scan --diff, and uploads the SARIF file with 0 findings upload.
name: 0 securityon: pull_request
permissions: contents: read security-events: write # upload to code scanning
jobs: scan: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 with: version: 0.41.0 # Pin the CLI fail-on: high env: OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}Use the key of the provider for the model you choose (see Models). The default model needs OPENAI_API_KEY.
What it does
Section titled “What it does”- Installs the release you pin.
versionis required. The installer downloads that release and refuses it unless its checksum matches the release’schecksums.txt. - Scans the change, not the repository. On a pull request it compares with the base branch and reviews only the files that differ and the code they connect to, so a check costs a fraction of a full scan. A FoxGuard pattern scan runs first and its hits are given to the scan as leads.
- Uploads the findings to code scanning with
0 findings upload. Alerts are matched by fingerprint, so pushing again updates them instead of duplicating them. - Fails the job when a finding is at or above
fail-on. The upload happens first, so the findings are visible either way.
Inputs
Section titled “Inputs”| Input | Default | Meaning |
|---|---|---|
version |
required | Release of 0 to install, for example 0.41.0. |
target |
. |
Folder to scan, relative to the workspace. |
diff-base |
the pull request’s base branch | Git ref to compare with. Outside a pull request, empty scans the whole folder. |
model |
the default model | Model for the scan thread. |
budget |
5 |
Stop at this many USD. The findings so far are kept and the job warns that they are partial. |
passes |
1 |
Scan this many times and merge the findings; the budget is split between them. |
fail-on |
none | Fail at or above this severity: critical, high, medium, low or info. |
upload |
true |
Upload to code scanning. Needs security-events: write. |
github-token |
the job’s token | Token used for the upload. |
Outputs: sarif (path of the file) and exit-code (0 done, 1 a finding at or above fail-on or the scan failed, 2 usage error, 3 stopped at the budget).
Things to know
Section titled “Things to know”- Same-repository pull requests only. A pull request from a fork has no access to your secrets, so it cannot run a model. Do not work around this with
pull_request_target: it would run a model on untrusted code with your key. - Use the default checkout. On a pull request it checks out the merge commit, so comparing with the base branch gives exactly the change.
- Code scanning needs a supported plan. On private repositories it is part of GitHub Code Security.
- Not a verdict. Findings are what a model found and checked in the code; the report says what it covered. Treat a clean run as “nothing found”, not “safe”.
Keep an SBOM and watch it for new advisories
Section titled “Keep an SBOM and watch it for new advisories”An SBOM and a regular advisory check are what the EU Cyber Resilience Act expects a manufacturer to keep. A scheduled job can regenerate the SBOM from your lockfiles and fail when a component has a known advisory; the lookup uses the public OSV database.
name: dependency advisorieson: schedule: - cron: "17 5 * * *" workflow_dispatch:
jobs: sbom: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - name: Install 0 run: | VERSION=0.41.0 # Pin the CLI RELEASE_BASE_URL="https://github.com/0sec-labs/0security/releases/download/v${VERSION}" INSTALL_DIR="$RUNNER_TEMP/0" sh <(curl -fsSL "https://raw.githubusercontent.com/0sec-labs/0security/v${VERSION}/install.sh") echo "$RUNNER_TEMP/0" >> "$GITHUB_PATH" - run: 0 sbom generate . -o sbom.cdx.json - run: 0 sbom check sbom.cdx.json - uses: actions/upload-artifact@v4 if: always() with: { name: sbom, path: sbom.cdx.json }0 sbom check exits 1 when it finds an advisory, so the job fails and you are told. This produces evidence for your own records; it does not by itself make a product compliant.
Collect evidence and draft vulnerability reports
Section titled “Collect evidence and draft vulnerability reports”For the EU Cyber Resilience Act (Regulation (EU) 2024/2847) a manufacturer keeps records of security testing and vulnerability handling, and reports actively exploited vulnerabilities to the authorities. 0 helps with the paperwork. It does not decide anything for you and never claims that a product complies.
0 findings evidence <folder> --repo . writes one folder: the findings with their triage state and timestamps (JSON and Markdown), a SARIF file, a CycloneDX SBOM from the lockfiles, the fix branches that exist in the repository, a manifest.json with the 0 version, date, commit and a SHA-256 for every file, and a README saying what each file shows and does not show. Add --check-advisories to look the SBOM up in the public OSV database (the only network use). Evidence and proof-of-concept bodies stay out unless you pass --include-evidence. To keep a record per release, run it in a job and upload the folder as an artifact:
- run: 0 findings evidence evidence --repo . --check-advisories - uses: actions/upload-artifact@v4 with: { name: security-evidence, path: evidence }The folder shows what was found, how it was triaged and which components are listed. It does not show that the code is free of other weaknesses, that a fix is correct or released, what is deployed, or that any obligation is met. Pull request links are not stored by 0, so fixes.md lists branches only.
When a finding is an actively exploited vulnerability, 0 findings report-draft <finding> --stage early-warning|notification|final writes the draft text for that stage. It takes the title, severity, location, status and timestamps from the finding and the rest from facts you pass (--product, --affected-versions, --exploited-since, --exploitation-evidence, --mitigation, --contact, --facts file.json). Anything you did not supply stays a visible [to be completed by the manufacturer: ...] blank, and nothing claims exploitation, user impact or a patch that you did not state. A person reviews the draft and submits it through the channel the authority designates; 0 sends nothing. Article 14 of the Regulation is the starting point: confirm the current requirements and deadlines with the authority and your advisers.