Coming soon ponytail
DietrichGebert ponytail/ponytail 81.8k DietrichGebert · @DietrichGebert ↗ Ponytail
Make your AI agent write only the code a task needs. A decision ladder gates every addition, and review and audit commands find over-engineering you shipped. Around 54% less code on one benchmark.
81.8k stars
Updated Jun 24, 2026 MIT 6 skills
The roster
Find Dead Code
DietrichGebert Scan your whole codebase for over-engineering and get a ranked list of what to delete, simplify, or replace with stdlib and native equivalents, biggest cut first.
ponytail/ponytail-audit 81.8k Coming soon DietrichGebert Coming soon ponytail-review
DietrichGebert ponytail/ponytail-review 81.8k Coming soon ponytail-debt
DietrichGebert ponytail/ponytail-debt 81.8k Coming soon ponytail-gain
DietrichGebert ponytail/ponytail-gain 81.8k Coming soon ponytail-help
DietrichGebert ponytail/ponytail-help 81.8k What it does
Ponytail makes your AI agent write only the code a task actually needs. Before it adds anything, it runs a short decision ladder: does this need to exist, is it already in the codebase, does the standard library or the platform already do it. The result is smaller diffs. On a real agentic benchmark editing a FastAPI and React repo, it cut about 54% of the code an unguided agent wrote, without dropping validation, error handling, security, or accessibility.
The bundled review, audit, and debt commands turn the same eye on code you have already shipped, finding the over-engineering to strip out. It installs as a plugin for Claude Code, Codex, and around 16 other agents.
When to use it
- Your agent writes more code than the task needs
A decision ladder gates every addition so you get the minimal change, not a new framework.
- You inherited an over-engineered codebase
Review and audit commands surface the abstraction and dead weight someone already shipped.
- You want the cleanup to be measurable
It tracks the debt removed and the reduction gained so the savings are visible, not vibes.
Common questions
What's included in Ponytail?
Ponytail is a set of six skills built around a decision ladder that gates every addition your agent makes, plus review and audit commands that catch over-engineering already in your code.
Is Ponytail free?
Yes. Ponytail is open-source under the MIT license with no paid tier. You install it once and it works across your agent's tasks.
Which agents does Ponytail work with?
Ponytail works with 16 agents, including Claude Code, Codex, and Copilot. In Claude Code you add it with /plugin marketplace add DietrichGebert/ponytail.
Does writing less code make the output less safe?
No. Ponytail keeps the safety guards in place while it trims code, and its own benchmark notes that a bare 'write one-liners' prompt drops a guard where Ponytail does not.