truecourse-ai/truecourse

What it solves

TrueCourse is an AI-powered architecture and code intelligence platform designed to prevent two primary types of software defects: code-level violations (such as security holes, performance bottlenecks, and architectural drift) and business-logic drift (where the actual implementation deviates from the project's documentation).

How it works

TrueCourse operates through two independent tools:

1. Analyze: This tool performs static and LLM-based analysis of the codebase. It uses tree-sitter AST visitors for over 1,200 deterministic rules (fast, no API calls) and 100 LLM-powered rules for deeper semantic analysis. It identifies issues across categories like security, bugs, architecture, and performance, storing results in a .truecourse/ directory for review via a CLI or dashboard.
2. Guard: This tool ensures the code matches the project's specifications. It scans documentation (PRDs, ADRs, READMEs, OpenAPI specs) to create a curated spec corpus. An LLM then generates declarative YAML scenario tests bound to specific sections of the documentation. These tests are executed deterministically (without an LLM in the loop) to detect when the code and documentation disagree.

Who it’s for

It is built for developers and engineering teams who want to automate the detection of architectural violations and ensure that their business logic remains synchronized with their technical documentation.

Highlights

• Hybrid Analysis: Combines fast deterministic rules with deep LLM semantic review.
• Spec-to-Test Automation: Automatically converts documentation into executable scenario tests to prevent business-logic drift.
• Deterministic Verification: Once Guard scenarios are generated, they run without LLM calls, making them suitable for CI/CD gates.
• Baseline Tracking: Uses a committed LATEST.json file to track violations and support diff-based analysis in pre-commit hooks.
• Broad Language Support: Supports JavaScript, TypeScript, Python, and C#.