dietrichgebert/ponytail

Ponytail is a plugin for AI coding agents that instructs them to write as little code as possible. The premise is that the best solution to most programming tasks is already built into the language, the platform, or a library you already have installed, and an AI agent will often write far more than necessary unless told not to. The plugin installs a short set of rules into the agent. Before writing any code, the agent is directed to check whether the thing already exists in the standard library, as a native browser or platform feature, or as an already-installed dependency. If none of those apply, it writes the minimum that actually solves the problem. The rules explicitly carve out exceptions: security, data handling, and accessibility are not cut. The README includes benchmark numbers from running five everyday tasks across three AI models with and without the plugin. The reported results show 80 to 94 percent fewer lines of code, 47 to 77 percent lower cost, and completions that finish three to six times faster compared to the same agent without any rules applied. The benchmark setup and raw numbers are included in the repository so anyone can reproduce them. Installation varies by which AI coding tool you use. For Claude Code and Codex, there are plugin install commands. For Cursor, Windsurf, Copilot, Aider, and Kiro, the repository includes the appropriate rules files that you copy into your project or home directory. A script is included to check that the rule text stays consistent across all those copies when you make changes. The project is released under the MIT license. It does not require any configuration file. The README is written in a consistent voice that matches the persona of a senior developer who considers unnecessary code a personal affront.