Developer Documentation

AI Guard Documentation

Explore technical guides, workflow configurations, and architectural deep-dives for catching unreliable and insecure AI-generated code.

Core Architecture & Philosophy

AI coding tools like GitHub Copilot, Cursor, and Claude Code generate impressive code rapidly, but probabilistic models do not guarantee correctness. Subtle bugs—like unhandled promise rejections, swallowed exceptions, or missing authorization checks—frequently pass standard syntax checks.

AI Guard provides a deterministic safety layer. Unlike LLM-based PR reviewers that hallucinate and add latency, AI Guard evaluates AST nodes with 100% predictability in milliseconds. Every rule is calibrated against hundreds of real AI code samples to ensure low false-positive rates.

Zero Type-Info Lag

AST structural heuristics run in <400ms without requiring slow TypeScript project compiler instantiation.

GitHub-Native SARIF

Upload findings straight into GitHub Code Scanning and display inline review annotations on PR diffs.

Prompt Shift-Left

Generate CLAUDE.md and .cursorrules files to stop models from generating anti-patterns at the prompt stage.