ERA OF MINDS

What is Ponytail? The Claude Code skill that codes like a lazy senior

By Era of Minds ·

Short answer

Ponytail is a free, MIT-licensed skill by Dietrich Gebert (DietrichGebert/ponytail / ponytail.dev) that makes Claude Code and other coding agents behave like a lazy senior: ship the smallest change that fully solves the task, reuse what is already in the repo or the standard library, and skip unrequested abstractions. Levels are lite, full (default) and ultra. Current release when we checked on October 18, 2026 was v5.3.2. The repo had about 166,393 stars and 9,001 forks. A viral X post about version 5 had about 214,200 impressions and 3,031 bookmarks at that same check.

Era of Minds cover: ‘Write less. Code.’ Ponytail 5.3 Claude Code lazy-senior skill, with the README before/after date-picker example.
Figure: the Ponytail 5 before/after date-picker example from the DietrichGebert/ponytail README (MIT): no skill builds 335 lines; Ponytail 5 reuses the repo Input with type="date" in 10 lines.

What is the Ponytail Claude Code skill?

You know the senior who has been at the company longer than the version control. You show them fifty lines. They say nothing, and replace them with one.

Ponytail puts that habit inside the agent. The core prompt lives in skills/ponytail/SKILL.md; a compact copy for rules-file hosts is AGENTS.md. The tagline on the README is blunt: "The best code is the code you never wrote."

It is not a model and not a hosted API. It is a ruleset plus plugin hooks that keep those rules active for the session. Dietrich's homepage at ponytail.dev brands the current generation as Ponytail 5.

How does Ponytail decide what to write?

After the agent understands the problem and the code the change touches, the skill runs a ladder. Stop at the first rung that holds:

  1. Does this need to exist at all?
  2. Is it already in this codebase?
  3. Does the standard library or a platform feature do it?
  4. Is there an installed dependency?
  5. Can it be one readable line?
  6. Only then: the minimum that works, plus one small test when the logic has a branch, loop, parser, money or security path.

The README is clear about what never gets cut: trust-boundary validation, data-loss handling, security and accessibility. Lazy about the solution, not about reading or finishing the callers and fixtures the change breaks. Deliberate shortcuts can be marked with a shortcut: comment so /ponytail-debt can collect them later.

How do you install Ponytail for Claude Code?

For Claude Code, the README asks for two separate prompts:

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

Codex uses the matching plugin commands, then you open /hooks, trust the two lifecycle hooks, and start a new thread. For a wide set of agents the skills CLI works:

npx skills add DietrichGebert/ponytail

That path installs the skills only. For the always-on ruleset, also add AGENTS.md or use a plugin. Step-by-step for Copilot CLI, Cursor, OpenCode, Gemini, Pi and others is in INSTALL.md. The README asks you to install only from DietrichGebert/ponytail on GitHub or @dietrichgebert/ponytail on npm.

What are Ponytail's levels and commands?

Intensity is a slash command:

  • /ponytail lite: build what was asked; name a smaller option in one line.
  • /ponytail or /ponytail full: the default ladder.
  • /ponytail ultra: also push back on any part of the request the need does not justify.
  • /ponytail off: turn it off for the session.

Review and audit tools:

  • /ponytail-review: review the current diff (or a named target such as staged, branch or a PR link) for bugs, security, real load, risky code without a test, slow paths and cuts. Each finding says what the code does, what goes wrong, how to fix it, and what happens if you skip it.
  • /ponytail-audit: the same checks for the whole repo, ranked by what to fix first.
  • /ponytail-debt: harvest deferred shortcut: comments into a ledger.
  • /ponytail-gain: show the measured impact scoreboard from the benchmark.
  • /ponytail-help: quick reference.

In Claude Code 2.1.290 or later, Ponytail 5.3 also shows a band above the prompt with cost, tokens, agent time, reuse rate and whether tests passed after the last change. Hide or show it with /ponytail band. A local board under ~/.config/ponytail/band/ can roll up past sessions.

What numbers does Ponytail 5 claim?

The README's Ponytail 5 benchmark (Claude Code, same agent with and without the skill, 39 tasks including a FastAPI + React repo, Opus 5.5, 5 runs each) reports roughly:

  • 53% less code, 41% less time, 26% less cost, 45% fewer tokens
  • 98% of risky logic ships with a test (68% without the skill)
  • On six security tasks (SQL injection, path traversal, forged tokens, rate limiting, malformed CSV, caching) it passed all 30 runs

A concrete before/after in the README: ask for a date picker. Without the skill the agent builds a calendar by hand (335 lines). With Ponytail 5 it reuses the repo's Input with type="date" (10 lines). Method and per-task tables live in the repo under benchmarks/results/.

An independent JetBrains AI blog retest of an earlier skill build, across 80 paired tasks, measured smaller but real savings: about 15% less code and 10.3% less cost, with no quality difference they could detect. Their write-up also notes the skill did not self-activate when only dropped into a skills folder; the plugin's SessionStart hook is what keeps the ruleset injected. Use the README numbers as the author's own benchmark, and expect the cut to be largest where the agent would otherwise over-build.

Who should try the Ponytail skill?

Anyone whose coding agent keeps shipping wrappers, new dependencies and "for later" options you did not ask for. If you already run Claude Code skills such as Taste-Skill (landing-page taste), earshot (audio for agents), Anatomy (isometric machines) or book-to-skill, Ponytail sits next to them as the always-on bias toward less code. Pair it with the Claude Code cost calculator if you want a sense of plan cost while the band tracks per-task burn.

When we checked on the morning of October 18, 2026 (Asia/Dubai), the repo showed 166,393 stars and 9,001 forks, created June 12, 2026, latest GitHub release v5.3.2 (published October 16 UTC). @bruceradley's October 16 post on X had about 214,198 impressions, 1,270 likes, 3,031 bookmarks and 60 reposts.

Key facts

  • Product: Ponytail, a lazy-senior ruleset / plugin for Claude Code and other coding agents. Smallest complete change; reuse first.
  • Author / repo: DietrichGebert/ponytail by Dietrich Gebert. Homepage ponytail.dev. MIT. Created June 12, 2026.
  • Stars / forks (checked Oct 18, 2026 morning Dubai): about 166,393 / 9,001.
  • Version: v5.3.2 (GitHub release, package.json and .claude-plugin/plugin.json).
  • X buzz: @bruceradley post Oct 16 ~214,198 impressions / 1,270 likes / 3,031 bookmarks / 60 RTs (checked Oct 18).
  • Claude Code install: two prompts, /plugin marketplace add DietrichGebert/ponytail then /plugin install ponytail@ponytail. Also npx skills add DietrichGebert/ponytail.
  • Commands: /ponytail [lite|full|ultra|off], /ponytail-review, /ponytail-audit, /ponytail-debt, /ponytail-gain, /ponytail-help, /ponytail band.

FAQ

What is Ponytail for Claude Code?

Ponytail is a free, MIT-licensed skill by Dietrich Gebert that puts lazy-senior rules into Claude Code and other coding agents. It pushes the agent to ship the smallest change that fully solves the task: reuse code already in the repo, prefer the standard library or a platform feature, and skip abstractions nobody asked for. Validation, error handling, security and accessibility stay off the chopping block.

How do I install Ponytail in Claude Code?

Send two separate prompts in Claude Code: /plugin marketplace add DietrichGebert/ponytail and then /plugin install ponytail@ponytail. The README warns that both prompts are required. You can also run npx skills add DietrichGebert/ponytail for the skills folder of many agents, or copy AGENTS.md / skills/ponytail/SKILL.md for hosts that only read a rules file. INSTALL.md covers Codex, Copilot CLI, Cursor, OpenCode, Gemini and more.

What do lite, full and ultra mean in Ponytail?

They are intensity levels. Lite builds what was asked and names a smaller option in one line. Full (the default) applies the full smallest-complete-change ladder. Ultra also questions the request itself before building. Switch with /ponytail lite, /ponytail full, /ponytail ultra or /ponytail off.

What commands does Ponytail add?

Besides /ponytail for levels, the README lists /ponytail-review (diff review for bugs, security, load, missing tests and cuts), /ponytail-audit (same checks for the whole repo, ranked), /ponytail-debt (harvest shortcut: comments into a ledger), /ponytail-gain (benchmark impact scoreboard) and /ponytail-help. Claude Code also gets /ponytail band to show or hide a cost/reuse band above the prompt.

How much code does Ponytail actually cut?

The DietrichGebert/ponytail README for Ponytail 5 reports a paired benchmark in Claude Code (39 tasks, Opus 5.5, 5 runs each): about 53% less code, 41% less time, 26% less cost and 45% fewer tokens versus the same agent with no skill, with 98% of risky logic shipping with a test (68% without). An independent JetBrains AI blog retest of an earlier skill build on 80 paired tasks measured smaller savings (about 15% less code and 10.3% less cost). Treat the big README numbers as the author's own benchmark, not a guarantee on every task.

Is Ponytail free and open source?

Yes. The DietrichGebert/ponytail repo is MIT-licensed. The homepage is ponytail.dev. Current release when we checked on October 18, 2026 was v5.3.2 (package @dietrichgebert/ponytail and .claude-plugin/plugin.json both say 5.3.2). The README asks you to install only from that GitHub repo or the matching npm package.

Who made Ponytail, and when did it go viral?

Dietrich Gebert (GitHub DietrichGebert). The repo was created on June 12, 2026. A widely shared X post by @bruceradley on October 16, 2026 (about version 5) had roughly 214,200 impressions, 1,270 likes, 3,031 bookmarks and 60 reposts when we checked on the morning of October 18. The repo had about 166,393 stars and 9,001 forks at that same check.

Sources

Keep reading