ponytail-gain Skill for Claude Code: Ponytail Benchmark Card

Shows a fixed scoreboard card of the Ponytail skill's published benchmark: code, tokens, cost and time against no skill.

Installs: 45.7K; on skills.sh
Stars: 182K; on GitHub
Updated: Oct 2026; last repo commit

ponytail-gain prints a scoreboard of the benchmark results the author publishes for his Ponytail skill. Type /ponytail-gain or ask what ponytail saves, and Claude draws ASCII bars for lines of code, output tokens, cost and time against the same agent with no skill.

The numbers are benchmark averages, not your repo. SKILL.md forbids printing any per-repo savings, because there is no real baseline to subtract. It is a one-shot display that changes no mode and writes no files, and it points to /ponytail-debt and /ponytail-audit for real figures about your own code.

Good for

People weighing whether to use the Ponytail skill
Ponytail users who want the benchmark figures at hand

Install ponytail-gain

Claude Code

Copy this prompt.

Install the ponytail-gain skill in Claude Code. Run this command:

npx skills add dietrichgebert/ponytail --skill ponytail-gain

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

The README gives the plugin as the Claude Code route; run its two commands as separate prompts. The plugin installs all six Ponytail skills plus hooks that run on session start and on every prompt, and it adds a status band to Claude Code. The skills CLI line installs only this display card, without /ponytail-debt or /ponytail-audit. Pick one route.

From the vendor's docs. Source.

What's in the skill

When it loads

Show ponytail's measured savings (code, cost, speed) from the benchmark. One-shot display. Use for /ponytail-gain, "what does ponytail save", "ponytail impact".

Sections in SKILL.md

Scoreboard
Honesty boundary
Boundaries

Key facts

Author: Dietrich Gebert, per the plugin manifest
Repo: dietrichgebert/ponytail
Installs on skills.sh: 45,691 (Nov 2, 2026)
GitHub stars: 182,027 (Nov 2, 2026)
License: MIT
Last commit: Oct 29, 2026
Files: 1 file including SKILL.md
Benchmark behind the card: Headless Claude Code with Opus 5.5 on 39 tasks, 5 runs each, geometric mean of per-task medians, 18 tasks with hidden checks
Part of: The Ponytail plugin: ponytail, ponytail-audit, ponytail-debt, ponytail-gain, ponytail-help and ponytail-review
skills.sh checks: Gen Agent Trust Hub pass, Socket pass, Snyk pass

What it includes

Renders the benchmark scoreboard once
Also triggers on "what does ponytail save" and "ponytail impact".
/ponytail-gain
Lines of code 47% of baseline, output tokens 55%, cost 74%, time 59%
Drawn as ASCII bars. Hidden checks passed: 97% against 96% with no skill. Tests where the logic needs one: 98% against 68%.
Scoreboard figures
Never prints a per-repo savings number
SKILL.md says the unbuilt version was never written, so a live repo has no baseline. Real per-repo figures come only from /ponytail-debt.
Honesty boundary
One-shot display; does not change mode, write flag files or persist anything
SKILL.md says it edits nothing and changes no mode.
Boundaries

From the official docs

Permissions

Low risk

What it can do

Read your data
Create and change things
Delete things
Read-only modeHow to turn it on

Watch out for

The plugin route installs hooks that write files and inject text
The manifest registers two hook files. They run Node scripts at session start, on subagent start and on each prompt. They write a mode flag per project, keep a copy of the status line script in your Claude config folder, and add rules plus a codebase map (built from `git ls-files`) to Claude's context at session start. Use the skills CLI line if you only want the card. github.com
The plugin asks Claude to edit your settings.json
At session start, if no status line is set, the hook adds text telling Claude to offer to add a statusLine entry to your Claude settings.json that runs the ponytail status line script. It nudges once and records that in a flag file. github.com
A task ledger is on by default
The plugin option "ledger" (default true) writes each task's cost, turns, time, tokens, reuse and tests to ~/.config/ponytail/band/<project>.csv after every turn and rebuilds a board.html there. Turn it off in the plugin settings if you do not want those files. github.com
/ponytail band import reads your past sessions
The README documents this command: it adds your past Claude Code sessions to the board. The script reads the session transcripts under ~/.claude/projects and writes rows to the same CSV files, in a detached background process. github.com
An optional webhook can send task rows out
If you set a webhook URL, each task row goes there as JSON after every turn, with task names unless you turn that off. The manifest says that with the URL empty, nothing leaves the machine. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	The skill alone is a one-shot display. SKILL.md says it edits nothing and changes no mode.
Talks to	Plugin route only: an optional webhook, empty by default, receives each task row as JSON
Scopes	The skill alone: renders text and edits nothing Plugin route: hooks run Node scripts at session start, subagent start and every prompt, and write small state files under ~/.claude and ~/.config/ponytail

Questions

How do I install ponytail-gain?

Use the install box above. The README's Claude Code route installs the whole Ponytail plugin, which includes hooks, so read the safety section first. The skills CLI line installs only this display skill.

Are these numbers about my own repo?

No. SKILL.md says they are benchmark averages, not your repo, and it forbids printing per-repo savings. For counted figures about your code, SKILL.md points to /ponytail-debt, a ledger of shortcuts you deferred, and /ponytail-audit.

Where do the figures come from?

From the author's own published benchmark of Ponytail 5: headless Claude Code with Opus 5.5 on 39 tasks, 5 runs each, against the same agent without the skill. SKILL.md names the source as benchmarks/results/2026-10-07-agentic.md in the repo and the README. They are the author's figures, not an independent test.

Do I need the main Ponytail skill?

SKILL.md only tells Claude to render the card. The commands it points to, /ponytail-debt and /ponytail-audit, are separate skills in the same repo, and the card describes what the main ponytail skill does.

Does ponytail-gain work in the Claude app or Claude.ai?

You can upload the skill folder as a zip under Customize, then Skills, in the Claude app. The README does not describe that route, so it is our generic instruction, not the author's. The plugin hooks and the band only apply to Claude Code.

Is it safe to use?

The skill alone is one Markdown file with no scripts, and its SKILL.md tells Claude to change nothing. The plugin route is different: it adds hooks that run Node scripts, write files under ~/.claude and ~/.config/ponytail, and can send task rows to a webhook you set. The permissions section lists each one.

Information

updated Nov 2, 2026

Repository: dietrichgebert/ponytail
Folder: skills/ponytail-gain

dietrichgebert/ponytail skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 9

ponytail · github.com
README.md · github.com
plugin.json · github.com
commands.md · github.com
SKILL.md · github.com
ponytail-activate.js · github.com
ponytail-mode-tracker.js · github.com
ponytail-band-import.js · github.com
register.tsx · github.com