RepoDoc: Intelligent Repo Auditing

Problem Statement: When developers join a new project or inherit a legacy codebase, onboarding is painfully slow. Manually auditing the architecture and searching for security vulnerabilities—like unencrypted state leaks, hardcoded credentials, or exposed .env files—costs an average of 8 to 16 hours of engineering time per repository. This manual process is not only a massive drain on operational budgets but also highly prone to human error, risking severe security breaches before the code even reaches production. The Solution: RepoDoc is an intelligent repository onboarding and security auditing tool powered by IBM Watsonx LLM architecture. It automates the entire codebase mapping and vulnerability detection process, condensing days of manual labor into mere seconds. Target Users & Interaction: The primary users are Software Engineers, Tech Leads, and Security Auditors. The interaction is completely frictionless. A developer simply inputs a public GitHub repository URL into the minimalist React UI. The FastAPI backend fetches the real file structure from the GitHub API and feeds it into the Watsonx client, instantly returning a comprehensive "Architecture Summary" and flagging critical "State Leaks." Creativity & Uniqueness: What makes RepoDoc truly unique is how it was engineered. It wasn't just built with AI; it was autonomously architected by IBM Bob 2.0. By leveraging Bob's /init command, I established persistent project rules in AGENTS.md. This allowed Bob to autonomously scaffold a strict Clean Architecture, write asynchronous HTTPX clients, and build a zero-dependency React frontend. Furthermore, the application features an isolated "mock boundary" for the Watsonx client, ensuring 100% testability, rate-limit immunity, and zero-friction local execution for evaluators. Effectiveness: RepoDoc effectively solves the codebase onboarding bottleneck by delivering enterprise-grade architectural insights and pinpointing security flaws at lightning speed, allowing developers to focus on building features rather than reading legacy structures.