Headroom

Ponytail for Codex: install and use the skill

Ponytail is usually described as a Claude Code skill, which leaves Codex users assuming it is not for them. It is. Ponytail works with the OpenAI Codex CLI too, and since Codex started metering by tokens it arguably matters more there. Here is how to turn it on and what changes.

Turning Ponytail on for Codex

  1. Download the Headroom app and open it once. It lives in the menu bar.
  2. Enable the Codex connector, which routes the Codex CLI through Headroom's local proxy.
  3. Open the Add-ons screen and toggle Ponytail on.
  4. Start a new Codex session. A running session will not pick up a newly installed skill.

The same toggle covers Claude Code, so if you switch between the two agents you configure this once.

Why this matters more on Codex now

Codex used to meter subscription usage by message count, which made response length nearly free: a 400-line answer and a 4-line answer cost the same. Since April 2026 that changed. ChatGPT Plus, Pro, and Business now meter Codex by tokens, so verbose output is charged for directly and drains the same 5-hour and weekly windows that your inputs do.

Ponytail's entire job is making the agent write less: no speculative abstractions, no scaffolding for later, no fifty lines where one will do. Under token metering, that converts straight into more work per window. See the Codex usage limits guide if you are regularly hitting the cap.

What changes in Codex's output

Ponytail installs a "lazy senior developer" posture: efficient, not careless. Before writing, the agent climbs a short 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 — and stops at the first rung that works. In practice you get smaller diffs, fewer new dependencies, and a short note on what was deliberately skipped instead of a page of design rationale.

What it does not do is cut corners on the things that matter. Input validation, error handling that prevents data loss, security, and accessibility are explicitly out of scope for simplification, and anything you ask for directly gets built in full.

Levels and commands

Three intensity levels, with full as the default: /ponytail lite, /ponytail full, /ponytail ultra. There are companion commands as well — /ponytail-review for an over-engineering pass on a diff, /ponytail-audit for a whole repo, and /ponytail-help if you just want the reference card. Running /ponytail-help in a fresh session is also the quickest way to confirm the skill loaded.

Pair it with input-side compression

Ponytail covers what Codex writes. It does nothing about what Codex reads, which on most sessions is the larger number: tool output, logs, files, and the previous turns being re-sent every time. Headroom compresses that side locally before the request leaves your machine, and it tracks your Codex subscription windows so you can see how much of the 5-hour and weekly limits you have left. Running both is the point — see how the pieces divide the work.

Try it on a real Codex task

Turn on the Codex connector and Ponytail, then run a job you have done before. Compare the diff size and the tokens it cost against last time.

Download for MacOS

7-day free trial · no credit card required