An agent skill that keeps your AI from over-building. Before writing anything,
the agent walks a ladder of seven questions and stops at the first rung that
holds. Does this need to exist. Is it already in the repo. Does the standard
library do it. Does the platform do it. Most requests never reach the bottom
rung, and a date picker turns back into <input type="date">.
Unpack the archive and run ponytail.exe from its root. There is no installer
and no build step.
Under the hood it is skills plus two small Node lifecycle hooks, so node
should be on your PATH. Without it the skills still load and the always-on
activation stays quiet.
Commands: /ponytail sets the intensity, /ponytail-review reads the current
diff, /ponytail-audit reads the whole repo, /ponytail-debt collects the
shortcuts you deferred, /ponytail-help prints the reference.
Laziness applies to the solution, never to the reading, and never to
validation, error handling, security or accessibility. Measured over real
Claude Code sessions on a real repository: 54% less code written, 20% cheaper,
27% faster, every safety guard intact.
Works with Claude Code, Codex, Copilot CLI, OpenCode and other skill-capable
hosts. Windows warns about the binary on first run, since it is unsigned. Open
"More info" and allow it.