Promptgenie

Static risk scanner and linter for AI-agent system prompts. Catch prompt injection vectors, overly permissive tool access, and trust boundary issues before they reach production — with CI-friendly output that fails builds on high-risk prompts.

What it does

• Static analysis of system prompts for prompt injection risk patterns
• Detects overly permissive tool access and broad instruction scopes
• Flags missing trust boundary controls between agent roles and data sources
• CI/CD integration — non-zero exit code on high-risk findings to fail builds
• JSON output for integration with security dashboards and SIEM pipelines
• Configurable risk thresholds and custom rule sets per project

Quick start

# Clone the repo
git clone https://github.com/mylesagnew/promptgenie
cd promptgenie

# Scan a system prompt file
./promptgenie scan path/to/system-prompt.txt

# Output JSON report
./promptgenie scan --format json path/to/prompt.txt

# Add to CI pipeline (fails on HIGH risk findings)
./promptgenie scan --threshold high prompts/

Related resources