Skip to main content
The PromptGuard Security Gate is a GitHub Action that runs automated red team tests against your security configuration on every pull request, ensuring security regressions are caught before merge.

​
Quick Start

# .github/workflows/security.yml
name: AI Security Gate
on: [pull_request]

jobs:
  security:
    runs-on: ubuntu-latest
    steps:
      - uses: promptguard/security-gate@v1
        with:
          api-key: ${{ secrets.PROMPTGUARD_API_KEY }}
          project-id: ${{ secrets.PROMPTGUARD_PROJECT_ID }}
          min-grade: B
          comment: true
          fail-on-regression: true

​
Inputs

InputRequiredDefaultDescription
api-keyYes—PromptGuard API key
project-idYes—PromptGuard project ID
api-urlNohttps://api.promptguard.coAPI base URL
min-gradeNoBMinimum acceptable grade (A, B, C, D, F)
fail-on-regressionNotrueFail if grade drops below baseline
commentNotruePost results as PR comment
budgetNo100Red team iteration count

​
Outputs

OutputDescription
gradeSecurity grade (A through F)
scoreNumeric score (0—100)
bypasses-foundNumber of bypasses discovered
reportFull JSON report

​
How It Works

  1. Calls the PromptGuard Red Team API with your project’s configuration
  2. Parses the response (grade, passed/failed vectors, score)
  3. Posts a PR comment with a summary table (if comment: true)
  4. Fails the check if grade is below min-grade
  5. Compares against baseline if fail-on-regression: true

​
PR Comment

When comment: true, the action posts a summary on the PR:
MetricValue
GradeB
Score84/100
Bypasses4
Block Rate92%

​
Using Outputs in Workflows

jobs:
  security:
    runs-on: ubuntu-latest
    steps:
      - id: gate
        uses: promptguard/security-gate@v1
        with:
          api-key: ${{ secrets.PROMPTGUARD_API_KEY }}
          project-id: ${{ secrets.PROMPTGUARD_PROJECT_ID }}

      - name: Check results
        run: |
          echo "Grade: ${{ steps.gate.outputs.grade }}"
          echo "Score: ${{ steps.gate.outputs.score }}"
          echo "Bypasses: ${{ steps.gate.outputs.bypasses-found }}"

​
Grading Scale

GradeBlock RateAssessment
A>= 95%Excellent security posture
B>= 85%Good, minor improvements possible
C>= 70%Acceptable, review failing test cases
D>= 50%Poor, significant gaps detected
F< 50%Critical, immediate action required

​
GitLab CI

# .gitlab-ci.yml
ai-security-gate:
  stage: test
  image: python:3.13-slim
  script:
    - pip install promptguard-sdk
    - python -c "
      import os, json, sys
      from promptguard import GuardClient
      client = GuardClient(api_key=os.environ['PROMPTGUARD_API_KEY'])
      result = client.redteam(project_id=os.environ['PROMPTGUARD_PROJECT_ID'], budget=100)
      print(json.dumps(result, indent=2))
      if result.get('grade', 'F') > 'B':
          sys.exit(1)
      "
  rules:
    - if: $CI_MERGE_REQUEST_ID

​
CircleCI

# .circleci/config.yml
version: 2.1
jobs:
  security-gate:
    docker:
      - image: cimg/python:3.13
    steps:
      - run:
          name: Run PromptGuard security gate
          command: |
            pip install promptguard-sdk
            promptguard redteam \
              --api-key "$PROMPTGUARD_API_KEY" \
              --format json \
              --preset default

​
Generic CLI (any CI)

For any CI system, use the PromptGuard CLI directly: 
# Install
curl -fsSL https://raw.githubusercontent.com/acebot712/promptguard-cli/main/install.sh | sh

# Run red team test and fail on grade below B
RESULT=$(promptguard redteam --api-key "$PROMPTGUARD_API_KEY" --format json)
GRADE=$(echo "$RESULT" | jq -r '.grade')

if [[ "$GRADE" > "B" ]]; then
  echo "Security grade $GRADE is below minimum (B). Failing build."
  exit 1
fi
echo "Security grade: $GRADE -- passed"

​
Best Practices

  1. Start with grade B: A reasonable minimum for most applications
  2. Enable regression detection: Catch security degradation early
  3. Run on every PR: Make security testing part of the development workflow
  4. Review PR comments: Understand which attack vectors pass through
  5. Combine with policy-as-code: Version your security config alongside your application

​
Next Steps

Red Team API

API reference for red team testing

CLI Tool

Run security tests from the command line

Policy-as-Code

Version your security configuration