kyontanaka/yagni-auditor

YAGNI stands for "You Aren't Gonna Need It," a software design principle warning against adding things before they are actually required. YAGNI Auditor applies that idea to Claude Code skills, plugins, and configuration files. You give it a CLAUDE.md file, a list of installed skills, or a configuration to review, and it analyzes the setup to find things that add cost or confusion without adding genuine value. The audit runs through four phases. The first looks for files and configurations that load on every interaction whether or not they are relevant, adding token usage quietly in the background. The second distinguishes between instructions that actually change how Claude behaves versus lists and catalogs that are just directories of information with no effect on behavior. The third checks whether multiple installed skills compete to handle the same request, which forces the AI to spend time deciding between them rather than doing the task. The fourth phase, added in version 2.0, detects activation cascades: situations where one skill triggers another which triggers a third, multiplying the processing involved for a single user request. The output is a scored report, zero to ten per architecture reviewed, sorted into critical issues, moderate issues, and things that look fine, with a post-audit checklist of concrete actions to take. The tool is a Claude Code skill, meaning it is a markdown file you copy into the Claude skills directory. Installation is a single command. There are no dependencies and the audit never runs external code, makes outbound network connections for data collection, or sends telemetry anywhere. Everything stays local. The README notes what this tool does not cover: it does not detect malware, does not analyze application source code, and does not do dynamic analysis that would require executing any code being reviewed. The project is licensed under MIT.