Brief: Code Context Infrastructure

A CLI that gives coding agents a deterministic understanding of how your code works. Brief helps Claude Code complete tasks 30% faster and use up to 45% fewer tokens. When an AI agent needs to modify something in your codebase, it searches around - grepping, reading files, building a mental model one piece at a time. Brief replaces that with a single query. You ask brief context "how does task creation work" and it locates key information, dynamically traces execution paths, and returns a layered understanding of where everything is and how it works. The agent starts with a complete picture of what exists and how it works, rather than spending time and tokens figuring out what is usually an incomplete or flawed understanding before doing the actual work. Brief provides telescoping context across the codebase, zoomed to the correct level for each relevant part: AST parsed manifest of every class, function, and dependency, traced execution flows through the call chain, LLM summaries of what each method and file does, detected conventions/contracts, and related code. I’ll demo the core loop, show how the static analysis and dynamic tracing work under the hood, and walk through what a real context package looks like at each layer - from high-level file descriptions down to traced execution paths with code. This is a working tool I use daily with Claude Code. I’ll show where it works well, where it breaks, and how I think it could be improved or built upon.