wonsukchoi/domain-experts

Domain Experts is an open source library that turns the knowledge of real job experts, such as lawyers or financial managers, into structured files an AI agent can load and reason with. Instead of asking an AI to simply act like a CFO or a contracts lawyer, which tends to produce a generic average of internet job descriptions, this project provides files containing the actual decision thresholds, warning signs, and worked examples a real practitioner would use. Each expert role is stored as a compact file containing the core reasoning a professional in that field would apply, plus optional reference files with deeper detail that only gets loaded when a task actually needs it, keeping token use efficient. The project argues that what makes these roles valuable is not just the text itself, since anyone could copy the files, but the process behind creating and correcting them: an authoring pipeline involving review passes, a scoring system, and automated checks that catch generic filler language before it is merged. Using the library involves installing it through a command line tool built on npm, either by matching a task description to the closest expert role and adding it, or by installing a lightweight router skill that automatically detects which expert a task calls for and loads that role's details on demand. The tool supports several AI coding assistants including Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Roo Code, and Amp, each of which can get a slash command that triggers this matching and loading process directly. The README specifically warns against pointing a tool's skill discovery directly at the raw roles folder, since it contains over two hundred separate files that would otherwise all get loaded into the system prompt at once. The project is released under the MIT license and documents its role authoring standards in a separate specification file for contributors who want to add new expert roles.