abhay-codes07/PromptGuard Red-Team
Red-team a live LLM chat endpoint against the OWASP LLM Top-10 and fail the build on a vulnerability regression. Optionally upload SARIF to code scanning.
View on GitHubTrust Signals
- Scorecard Score
- not yet scored
- Maintenance Recency
- Activelast commit Jun 19, 2026
- License
- MIT
Pinned Snippet
uses: abhay-codes07/promptguard-mcp@03c268bc50c5283dce4ab033c7402f1466f13df2 # no releases — HEAD as of 2026-07-16tags can be moved; commit SHAs can't. why a SHA?
Inputs
| name | description | required | default |
|---|---|---|---|
| url | Target chat endpoint URL (receives a JSON POST). | yes | — |
| auth | Value for the Authorization header, e.g. 'Bearer ${{ secrets.TARGET_TOKEN }}'. | no | "" |
| response-path | Dotted path to the assistant text in the JSON response. | no | choices.0.message.content |
| max-score | Fail the build (exit 3) if any OWASP category score exceeds this 0-100 threshold. Higher score = more vulnerable. | no | 30 |
| max-attacks | Cap the run at N attacks. Empty = full corpus. | no | "" |
| concurrency | Max concurrent in-flight attacks. | no | 5 |
| formats | Comma-separated report formats: json, markdown, sarif. | no | json,markdown,sarif |
| out-dir | Directory to write reports into. | no | promptguard-reports |
| upload-sarif | Upload the SARIF report to GitHub code scanning. | no | true |
Outputs
no outputs