Skip to content

pre-commit.md

Latest commit

 

History

History
169 lines (123 loc) · 3.61 KB

pre-commit.md

File metadata and controls

169 lines (123 loc) · 3.61 KB

Pre-commit Hook Integration

Prompt Shield provides pre-commit hooks that scan staged files for prompt injection patterns and PII before every commit. This catches security issues early, before they enter your repository.

Available Hooks

Hook ID Description
prompt-shield-scan Scans for prompt injection patterns (role hijacking, instruction override, etc.)
prompt-shield-pii Scans for PII (emails, phone numbers, SSNs, credit card numbers, etc.)

Setup

1. Install pre-commit

pip install pre-commit

2. Add to your .pre-commit-config.yaml

repos:
  - repo: https://github.com/mthamil107/prompt-shield
    rev: v0.3.0
    hooks:
      - id: prompt-shield-scan
      - id: prompt-shield-pii

3. Install the hooks

pre-commit install

Now Prompt Shield will automatically scan every staged text file when you run git commit.

Configuration

Threshold

Adjust the confidence threshold for prompt injection detection (default: 0.7). Lower values are more sensitive; higher values reduce false positives.

repos:
  - repo: https://github.com/mthamil107/prompt-shield
    rev: v0.3.0
    hooks:
      - id: prompt-shield-scan
        args: ["--threshold", "0.8"]

Excluding Files

Skip files that match glob patterns:

repos:
  - repo: https://github.com/mthamil107/prompt-shield
    rev: v0.3.0
    hooks:
      - id: prompt-shield-scan
        args: ["--exclude", "*.min.js", "--exclude", "vendor/*"]
      - id: prompt-shield-pii
        args: ["--exclude", "tests/fixtures/*"]

You can also use the built-in exclude key from pre-commit:

hooks:
  - id: prompt-shield-scan
    exclude: "^(vendor/|node_modules/)"

Using Only One Hook

You do not need to enable both hooks. Pick the ones relevant to your project:

# Only scan for prompt injection
hooks:
  - id: prompt-shield-scan
# Only scan for PII
hooks:
  - id: prompt-shield-pii

How It Works

Prompt Injection Hook

The prompt-shield-scan hook runs each staged text file through the PromptShieldEngine with a lightweight configuration optimized for speed:

  • Vault, feedback, canary, and history subsystems are disabled
  • The semantic classifier (which requires ML model downloads) is disabled
  • Only fast regex-based and heuristic detectors run

Files that trigger any detector above the confidence threshold cause the commit to be blocked.

PII Hook

The prompt-shield-pii hook uses the PIIRedactor to scan for common PII patterns including:

  • Email addresses
  • Phone numbers
  • Social Security Numbers
  • Credit card numbers
  • IP addresses
  • AWS keys and other API credentials

Any PII detection blocks the commit.

Output

Clean files show green PASS output:

PASS src/app.py
PASS config/settings.yaml

Detections show red FAIL output with details:

FAIL prompts/system.txt — 2 detection(s), risk score 0.85
  [HIGH] d001_instruction_override (confidence: 0.82)
    Detected instruction override pattern
    Line 5: 'ignore all previous instructions'

Running Manually

Run the hooks against all files without committing:

pre-commit run prompt-shield-scan --all-files
pre-commit run prompt-shield-pii --all-files

Run against specific files:

pre-commit run prompt-shield-scan --files src/prompts/*.txt

CI Integration

Add to your CI pipeline to scan all files on every push:

# GitHub Actions example
- name: Run Prompt Shield hooks
  run: |
    pip install pre-commit
    pre-commit run prompt-shield-scan --all-files
    pre-commit run prompt-shield-pii --all-files