← Discover
Discover · GitHub · Ponytail

Rule Set for AI Coding Agents

An MIT-licensed rule set and plugin system for AI coding agents. Its aim is to preserve validation, error handling, security, and accessibility while writing only the code required for a task.

Updates

  • September 5, 2026: Stars 119,108 → 126,773, latest release v4.9.0 (August 7, 2026).
  • September 1, 2026: Stars 113,377 → 119,108, latest release v4.9.0 (August 7, 2026).
  • August 27, 2026: Stars 110,483 → 113,377, latest release v4.9.0 (August 7, 2026).

Installation

Add the Claude Code marketplace
/plugin marketplace add DietrichGebert/ponytail
Install the Claude Code plugin
/plugin install ponytail@ponytail

Running it

Select the Ponytail level
/ponytail full
Start diff review
/ponytail-review

What does this tool do?

The rule ladder is applied after reading the code affected by a change. The adjusted agentic benchmark reported, on 12 tasks in a real FastAPI and React repository with Haiku 4.5 versus a no-skill baseline, an average of 54% fewer lines of code, 22% fewer tokens, 20% lower cost, and 27% shorter time. These results are limited to specific test conditions.

Who it is for

Those who want to add validation, security, and accessibility rules to coding workflows on Claude Code, Codex, Gemini CLI, and other supported agent hosts.

What not to expect

Not for generalizing specific benchmark results to all projects or for applying critical production changes without human review.

Highlights

  • Task-focused rules aimed at reducing unnecessary code
  • A review approach that preserves validation, error handling, security, and accessibility
  • Plugins or instruction adapters for Claude Code, Codex, Gemini CLI, and other hosts

First-use flow

  1. Install the Ponytail integration for your agent host
  2. Verify that the installation is active inside the host
  3. Choose the appropriate Ponytail level
  4. Run review or audit flows on changes

Safe start

Percentages are the adjusted agentic benchmark averages on 12 tasks in a real FastAPI and React repository, with Haiku 4.5 and n=4. A separate adversarial tier reported 100% safety. The previous single-shot 80–94% range is not a general average.

First task prompt

🤖 Paste this into your AI agent (Claude Code · Codex · Antigravity)

Write only the code required by the task, then review changes for validation, error handling, security, and accessibility.

Related dictionary terms

BenchmarkAgenticTokenAgentCLIArtificial Intelligence

Links

TreScout did not build this tool · we found it in GitHub trends and wrote it up. This page describes the repository as of 2026-08-25: The star count and our text belong to that day, the repository may have changed since. Check the repository link for the current state. This page was machine-translated from the Turkish original · the Turkish version prevails.