Scan system prompts for missing defenses against 12 attack vectors. Runs in your CI pipeline — blocks PRs with weak prompts before they reach production.

Pure regex. Zero LLM cost. < 5ms per file.

- uses: ppcvote/prompt-defense-audit-action@v1
  with:
    path: "prompts/**/*.txt"
    min-grade: B

That's it. The action scans all matching files, posts a PR comment with results, and fails the check if any file scores below the minimum grade.

The action posts a summary table on your PR:

🛡️ Prompt Defense Audit

❌ 1 of 3 file(s) below minimum grade B

File	Grade	Score	Defended	Missing defenses
prompts/support.txt	🔴 F	17/100	2/12	role-escape, data-leakage, indirect-injection +7 more
prompts/chatbot.txt	🔵 B	75/100	9/12	unicode-attack, context-overflow, abuse-prevention
prompts/admin.txt	🟢 A	92/100	11/12	unicode-attack

The comment is updated on each push (no spam).

name: Prompt Security
on: [pull_request]

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: ppcvote/prompt-defense-audit-action@v1
        with:
          path: "**/*.prompt.txt"
          min-grade: C

- uses: ppcvote/prompt-defense-audit-action@v1
  with:
    path: "config/prompts/**/*.txt"
    min-grade: A

- uses: ppcvote/prompt-defense-audit-action@v1
  with:
    path: "prompts/**/*.txt"
    min-grade: B
    sarif: results.sarif

- uses: github/codeql-action/upload-sarif@v3
  if: always()
  with:
    sarif_file: results.sarif

- uses: ppcvote/prompt-defense-audit-action@v1
  id: audit
  with:
    path: "prompts/**/*.txt"

- run: echo "Lowest grade: ${{ steps.audit.outputs.lowest-grade }}"

- uses: ppcvote/prompt-defense-audit-action@v1
  continue-on-error: true
  with:
    path: "prompts/**/*.txt"
    min-grade: A

Based on OWASP LLM Top 10:

1. Globs for prompt files matching your pattern
2. Scans each file with prompt-defense-audit (deterministic regex, no LLM)
3. Posts a markdown summary table as a PR comment
4. Optionally outputs SARIF for GitHub Code Scanning integration
5. Fails the workflow if any file is below the minimum grade

These three layers are complementary: static prompt → code flow → behavioral testing.

MIT — Ultra Lab

• prompt-defense-audit — The underlying scanner (npm package + CLI)
• OWASP LLM Top 10