Repo Dev Tools

DietrichGebert/ponytail

A single prompt/skill that makes AI coding agents write only the code a task needs, reusing existing code first, with review, audit and debt commands.

  • 160k GitHub stars
  • JavaScript
  • ⚖️ MIT
  • 🎯 Beginner
View on GitHub (opens in a new tab) Setup guide
DietrichGebert/ponytail preview image

What it is

Ponytail is one prompt (skills/ponytail/SKILL.md, with a compact AGENTS.md version) that is loaded into AI coding agents so they write only what the task needs. Before writing code, the agent climbs a ladder: does it need to exist, is it already in the codebase, does the standard library or a native platform feature cover it, then an installed dependency, one line, and only then the minimum that works. It never cuts validation, error handling, security or accessibility, and leaves one small test behind for risky logic. It also adds review, audit, and debt-tracking commands.

Who it's for

  • Developers using AI coding agents such as Claude Code or Codex who want smaller, simpler generated code
  • Teams who want AI-generated changes reviewed for bugs, security, load and missing tests, not just trimmed
  • Users of other agents (Copilot, Cursor, OpenCode, Gemini and others) who can load a rules file or skill

Requirements

Requirements

  • A supported AI coding agent (README says it works with 20 agents)
  • A skill-capable host (Claude Code, Codex, Devin CLI, OpenCode, Gemini, pi, Hermes Agent, Qoder, Grok Build) to use the slash commands
  • Instruction-only agents (Cursor rule file, Windsurf, Cline, Copilot, Kiro, Antigravity) get the always-on ruleset without the commands

Setup

  1. Install in Claude Code

    Run these as two separate prompts.

    bash
    /plugin marketplace add DietrichGebert/ponytail
/plugin install ponytail@ponytail

  2. Install in Codex

    Then open /hooks in Codex, trust its two lifecycle hooks, and start a new thread.

    bash
    codex plugin marketplace add DietrichGebert/ponytail
codex plugin add ponytail@ponytail

  3. Any other agent

    Copy AGENTS.md into your project, or ask your agent to install skills/ponytail/SKILL.md as a skill. Step-by-step instructions for Copilot, Cursor, OpenCode, Gemini and others are in INSTALL.md.

Examples

Set the intensity level

Prompt
prompt
/ponytail [lite | full | ultra | off]

Expected output: Sets the intensity or turns it off. With no argument it switches on at the default level if off, otherwise reports the current level.

Review the current diff

Prompt
prompt
/ponytail-review

Expected output: Reviews like the senior dev who gets paged when it breaks: bugs, security, real load, risky code without a test, slow paths, and what to cut. You can name a target such as uncommitted, staged, branch, or a PR link.

Audit the whole repo

Prompt
prompt
/ponytail-audit

Expected output: Runs the same checks across the whole repo, mapping entry points and data flow first, then ranking findings with what to fix first.

Harvest deferred shortcuts

Prompt
prompt
/ponytail-debt

Expected output: Collects the shortcut: comments the agent left into a ledger so deferred work isn't forgotten.

Invoke a command in Codex

Prompt
prompt
$ponytail:ponytail-review

Expected output: In Codex CLI and the IDE extension, the commands are skills under the plugin's namespace and are invoked this way.

Pros & cons

Pros

  • Pro:Reported benchmark (39 tasks, 5 runs each, Claude Code): -53% code, -41% time, -26% cost, -45% tokens versus the same agent without the skill
  • Pro:Keeps safety: validation, error handling, security and accessibility are never cut, and logic gets a small test (98% of risky logic shipped with a test vs 68% without)
  • Pro:Works across many agents via a single prompt, with an AGENTS.md fallback and an MIT license
  • Pro:Includes review, audit and debt-ledger commands beyond the core prompt

Cons

  • Con:Slash commands require a skill-capable host; instruction-only adapters get just the always-on ruleset
  • Con:Cursor with hooks only supports /ponytail level switching, typed as a plain message
  • Con:Benchmark figures come from the author's own tests, using Claude Code with Opus 5.5

Images

DietrichGebert/ponytail preview image
DietrichGebert/ponytail preview image · DietrichGebert/ponytail on GitHub · MIT
DietrichGebert/ponytail preview image