Platform·GitHub Action
QWED checks in your own CI workflows.
A GitHub Action that scans for leaked secrets, unsafe code and dangerous shell scripts on the runner, and can send model output to a QWED API to be checked.

At a glance
- Use
- QWED-AI/qwed-verification@v9.1.0
- Listed on
- GitHub Marketplace, as QWED Protocol Verification
- Output
- Text, JSON or SARIF
- Data
- Scan modes never leave the runner
- Licence
- Apache-2.0
Modes
- scan-secrets
Leaked secrets
API keys, tokens and SSH keys.
- scan-code
Unsafe code
eval(), exec(), subprocess and unsafe imports.
- verify-shell
Dangerous scripts
curl | bash, rm -rf and sudo escapes.
- verify
One claim or one piece of code
Math and logic check the query input; code checks the llm_output input. A QWED API runs the check.
Inputs
| Input | Default and use |
|---|---|
| action | verify. One of scan-secrets, scan-code, verify-shell or verify. |
| paths | Comma-separated globs for the scan modes. Unset, each mode uses its own: **/*.py for scan-code, **/*.sh for verify-shell, and env, JSON, YAML, TOML and INI files for scan-secrets. |
| output_format | text. Or json, or sarif, which writes qwed-results.sarif. |
| fail_on_findings | true. Fails the step on any finding, and on a scan that is not admitted. |
| engine | math. For verify: math, logic or code. |
| query | For verify with math or logic: the claim to check. |
| llm_output | For verify with code: the code to check. |
| api_key | For verify. Optional. |
Outputs
| Output | What it holds |
|---|---|
| verified | true only when the check passed, or the scan covered its files and found nothing. |
| findings_count | Scan modes: the number of findings. |
| files_scanned | Scan modes: the number of files scanned. 0 means the scan was not admitted. |
| explanation | verify: the proof, or why it did not hold. |
| sarif_file | With output_format: sarif, the path of the SARIF file. |
| badge_url | A shields.io badge for the result. |
Get started
name: QWED Verificationon: [pull_request]jobs: verify: runs-on: ubuntu-latest steps: - uses: actions/checkout@v4 - uses: QWED-AI/qwed-verification@v9.1.0 with: action: scan-secrets paths: "**/*.env,**/*.json" output_format: sarif fail_on_findings: "true"Good to know
- scan-secrets, scan-code and verify-shell run entirely inside the runner.
- A scan that matches no files, or cannot read a file it matched, is not admitted. With fail_on_findings left at true, the step fails.
- verify sends the check to http://localhost:8000 from inside the action container, where localhost is the container itself. A QWED API on the runner or in a service container is not reached that way, and the action has no input to change the address yet.