The Problem

Every time you start an AI coding session — Claude Code, Cursor, Codex, Cline — the agent re-discovers your project from scratch. It reads random files, guesses at your architecture, and burns tokens figuring out what you already know.

The result? Worse outputs. Wasted money. Slower development.

The Fix

npx repobrief init

RepoBrief scans your repo and auto-generates a .repobrief/ context directory:

| File | What it contains | |------|-----------------| | architecture.md | Project structure, entry points, build system | | patterns.md | Coding conventions detected from your actual code | | dependencies.md | Key deps categorized as runtime vs dev | | hotfiles.md | Most-changed files from git history — where bugs live | | context.json | Machine-readable version for tool integrations |

Export to Your AI Tool

repobrief export --format claude    # → CLAUDE.md
repobrief export --format cursor    # → .cursorrules
repobrief export --format codex     # → AGENTS.md
repobrief export --format markdown  # → universal summary

One command. Your AI agent gets a complete briefing before writing a single line of code.

Keep It Fresh

repobrief update

Re-scans your project and updates the context. Fast enough to run on every commit.

Why RepoBrief?

| Without RepoBrief | With RepoBrief | |---|---| | AI re-reads your entire project every session | AI gets a structured briefing in seconds | | Context files go stale immediately | repobrief update keeps them current | | One format locked to one tool | Export to any AI tool's format | | You write context docs by hand | Auto-generated from real code analysis | | Doesn't scale past small projects | Built for monorepos and large codebases |

Supported Projects

RepoBrief auto-detects your stack:

• JavaScript / TypeScript — package.json, tsconfig.json
• Python — requirements.txt, pyproject.toml, setup.py
• Swift / Xcode — Package.swift
• Rust — Cargo.toml
• Go — go.mod

Install

# Run directly (no install needed)
npx repobrief init

# Or install globally
npm install -g repobrief

How It Works

1. Structure Analysis — Detects project type, framework, build system, key directories, and entry points
2. Dependency Analysis — Parses manifest files, categorizes runtime vs dev dependencies
3. Git History Analysis — Identifies high-churn files, recent activity, and contributor patterns
4. Pattern Detection — Scans your code for naming conventions, import styles, and error handling patterns
5. Context Assembly — Combines all analysis into structured markdown + JSON
6. Export — Transforms context into the format your AI tool expects

Contributing

RepoBrief is open source under the MIT License. Issues, PRs, and feedback welcome.

License

MIT