Ponytail is a ruleset and plugin that cuts AI-generated code down to what’s strictly needed. In benchmarks on the Claude API—across Haiku, Sonnet and Opus models—it delivered 80–94% less code, ran 3–6× faster and cost 42–75% less than an unskilled agent. It does this by walking a simple ladder before writing anything: skip if YAGNI, use stdlib or native features first, pull in dependencies only when essential, then boil whatever remains into a single line if possible. Validation, error handling, security checks and accessibility never get trimmed away. Under the hood, ponytail injects its rules into every agent turn. You install it via plugin commands— for example, “/plugin marketplace add DietrichGebert/ponytail” in Claude Code or “codex plugin install ponytail@ponytail” in Codex’s desktop app—then trust two tiny Node.js hooks. Once active, you get slash commands like “/ponytail full” to set intensity, plus “/ponytail-review”, “/ponytail-audit” and “/ponytail-debt” for diff reviews, repo audits and tracking deferred shortcuts. It also works with Copilot CLI, OpenCode, Gemini/Antigravity CLI, ClawHub and other agent hosts by copying the same rules files into their config paths. The plugin covers both “lite” and “ultra” modes: lite enforces minimal trimming, full applies the ladder strictly, and ultra unleashes every shortcut. You can set a default with the PONYTAIL_DEFAULT_MODE environment variable or a field in your config.json. For editor integrations—Cursor, Windsurf, Cline, GitHub Copilot, Kiro—you just copy matching .rules or .md files from the repo into each tool’s steering or instructions directory. No extra setup beyond installing the plugin and placing rule files, yet every session gains consistent “pony­tailed” guidance on avoiding over-engineering.