PromptShield Action

GitHub Action for integrating PromptShield CLI into CI/CD pipelines — effectively an ESLint for prompts. It scans repositories for prompt injection risks, insecure AI patterns, jailbreak attempts, and unsafe LLM configurations before they reach production. 🛡️

Features

• ESLint for Prompts Shift AI security left. Detect prompt vulnerabilities during CI, not after deployment.

• Two Execution Modes

  • Gatekeeper Mode → Fast, fail-fast security checks for CI enforcement.
  • Audit Mode → Full repository analysis with Markdown reports and optional PR comments.

• PR-Aware Scanning Automatically scans only changed files for pull requests and pushes when using scan-mode: auto.

• Developer-Friendly Reports Generates actionable findings with severity levels and remediation guidance.

---

Usage

Gatekeeper Mode ⚡ (Fast Fail)

Best for strict CI enforcement.

Gatekeeper mode uses --check internally and exits immediately when findings match the configured severity threshold.

name: Security Scan

on: [push, pull_request]

permissions:
  contents: read
  pull-requests: write

jobs:
  promptshield:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4

      - name: Run PromptShield
        uses: promptshield-io/action@v1
        with:
          report: 'false'

---

Audit Mode 📋 (Detailed Reports + PR Comments)

Best for security reviews and pull request workflows.

Audit mode performs a complete scan, generates .promptshield/workspace-report.md, and can automatically post findings as a PR comment.

name: Security Scan

on: [pull_request]

jobs:
  promptshield:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v4

      - name: Run PromptShield Audit
        uses: promptshield-io/action@v1
        with:
          report: 'true'
          token: ${{ secrets.GITHUB_TOKEN }}

---

Inputs

Input	Description	Default
patterns	Glob patterns specifying files to scan.	**/*
min-severity	Minimum severity threshold to report (LOW, MEDIUM, HIGH, CRITICAL).	LOW
no-inline-ignore	Disables inline suppression comments.	true
report	Enables Audit mode when true. Uses Gatekeeper mode when false.	true
scan-mode	Scan strategy: auto, diff, or full. auto scans changed files for PRs/pushes when possible.	auto
token	GITHUB_TOKEN used for posting PR comments in Audit mode.	null

---

Recommended Setup

For most repositories:

with:
  report: 'true'
  scan-mode: 'auto'
  min-severity: 'MEDIUM'

This provides:

• Fast incremental scans
• Actionable reports
• Reduced CI noise
• Reasonable security enforcement

---

Why PromptShield?

Traditional security scanners understand code. PromptShield understands LLM attack surfaces.

It helps detect:

• Prompt injection vectors
• Jailbreak instructions
• Dangerous system prompt overrides
• Unsafe tool invocation patterns
• Risky AI configurations
• Hidden prompt manipulation attempts

Because "ignore previous instructions" should never reach production 😄