ponytail

Invoke the ponytail skill, which applies a decision ladder that questions whether the task is needed, reuses existing code, and prefers standard library solutions before writing new code. The default full mode enforces the shortest working diff on every response.

Run /ponytail lite, /ponytail full, or /ponytail ultra to switch modes mid-session. Lite suggests lazier alternatives while building what was asked, full enforces the ladder by default, and ultra aggressively deletes and challenges requirements.

Do not simplify away input validation at trust boundaries, error handling that prevents data loss, security measures, accessibility basics, or anything explicitly requested. The skill also requires fully understanding the problem and tracing all affected code before choosing a minimal fix.

Non-trivial logic such as branches, loops, parsers, or money and security paths must ship with one runnable check, like an assert-based self-check or a single small test file. Trivial one-liners need no test, and test frameworks or fixtures are avoided unless requested.

It requires grepping every caller of the function being changed and fixing the root cause in the shared code path rather than patching the reported symptom. This produces a smaller diff while fixing all sibling callers at once.