dsh-ds-ponytail

AI coding assistants often introduce unnecessary dependencies, rewrite modules, or layer on defensive logic, inflating codebases and maintenance costs. Ponytail plugs explicit minimal-solution guidance into DeepSeek Harness, steering the model to reuse existing code, standard libraries, and platform features while making the smallest changes that preserve behavior.

The bundle registers six load-on-request skills: ponytail simplifies a coding task at lite, full, or ultra intensity; ponytail-review flags removable complexity in a diff; ponytail-audit scans a repository for behavior-preserving simplifications; ponytail-debt collects shortcut comments into a debt ledger; ponytail-gain surfaces the upstream benchmark scoreboard with its model and workload caveats; and ponytail-help lists available levels and skills. A browser module also adds a Ponytail page under Settings to confirm loading and verify host-side skill availability by name.

Best for teams and solo developers who want to keep AI-generated code lean, track simplification debt, and make precise changes within an existing architecture. It does not alter Harness session mechanics or persist mode state; every activation is one-shot and reversible.