ponytail cuts up to 94% of the code in Claude Code sessions
The ponytail skill pushes an AI agent to write as little code as possible. Before it implements anything, the agent climbs a ladder of checks (is this even needed, does it already exist in the codebase, does a standard library or a native platform function already solve it) and stops at the first step that fits.
The benchmark ran on real work: a headless Claude Code session editing the full-stack-fastapi-template repo, 12 tasks, the same agent with and without ponytail, n=4, on Haiku 4.5. Against the baseline, ponytail produced 54% less code on average, 22% fewer tokens, 20% cheaper, and 27% faster.
The 94% peak shows up on rebuild tasks: a date picker shrank from 404 to 23 lines, a color picker from 287 to 23. You install it in Claude Code with marketplace add DietrichGebert/ponytail, MIT license.
Up next
UST puts Claude Code to work reading chip pinouts and writing regression tests
UST engineers used to hand-script chip validation and run it in four-day cycles. Now Claude Code reads pinouts and schematics itself, writes regression tests, and checks live data against a digital twin. Per UST, its iDEC pipeline already cuts cycles by 50% to 70%.
Related stories
- Context compression cut 39% of tokens and saved almost nothing
- ponytail’s token savings: 10.3% off the bill, not the promised 20%
- A Claude Code mod that reverts its own failed ideas
- Claude Code shipped its own TypeScript in npm source maps
- Show HN: Generate a Claude Code or Codex agent fleet from one HTML file
- pairmark races Claude Code against Codex in your repo
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.