Install with the Headroom app
- Download the Headroom app and drag it to your Applications folder.
- Open it once. It sets itself up and lives in the menu bar.
- Open the Add-ons screen and toggle Ponytail on.
- Start a new Claude Code session. Existing sessions do not pick up a newly installed skill.
Because Headroom manages the install, you also get the same add-on applied to Codex from that one toggle, and updates land without you doing anything.
Installing Ponytail by hand
Ponytail is an ordinary Claude Code skill, so it lives in your skills directory like any other. If you manage your Claude Code setup in a dotfiles repo and want Ponytail versioned alongside it, place the skill under ~/.claude/skills/ and restart Claude Code so it picks up the new directory.
This is the right choice if you already curate your own skills and want full control over the version you run. The trade-off is that updates are now your job: there is no auto-update, and a skill that drifts a few versions behind quietly loses the improvements. The app path exists because most people would rather not track that.
Check that it is actually on
Start a fresh Claude Code session and run /ponytail-help. If the skill is installed you get a reference card listing the modes and commands. If nothing happens, the skill directory is not where Claude Code is looking, or the session predates the install.
The behavioural check is more convincing: ask for something easy to over-build, like a cache in front of a function. With Ponytail active the agent should reach for the smallest thing that works and tell you what it skipped, instead of handing you a configurable cache class you did not ask for.
Intensity levels
Ponytail has three levels, and full is the default. Switch at any time with /ponytail lite, /ponytail full, or /ponytail ultra. Full enforces the whole ladder: does this need to exist at all, is it already in the codebase, does the standard library or a native platform feature cover it, can it be one line. Lite is a lighter touch for exploratory work; ultra is for when you want the agent to be genuinely ruthless about cutting code.
There are companion commands too: /ponytail-review reviews a diff purely for over-engineering, /ponytail-audit does the same across a whole repo, and /ponytail-debt collects the shortcuts Ponytail deliberately left behind so they get tracked instead of forgotten.
What Ponytail does and does not cover
Ponytail works on the output side of the bill: the code the agent writes back. That is real money on code-generation work, and it compounds, because over-built code costs again every time the agent re-reads and edits it. What Ponytail does not touch is the input side — the bloated tool output, logs, and boilerplate piling up in your context window. That is Headroom's job, and running both is the point. See Headroom vs Ponytail vs RTK for how the pieces divide the work.