Dietrich Gebert publishes Ponytail, a set of instructions and review tools for AI coding agents that emphasizes reusing existing code and implementing only what a task needs.[1][2]
Start with Ponytail →
Links
GitHub Contributions
Ponytail asks an agent to inspect the surrounding code before choosing a solution. Its solution ladder checks whether a feature is needed, then looks for existing code, standard-library functions, platform capabilities and installed dependencies before writing a new implementation. Review and audit commands examine bugs and risk alongside opportunities to simplify.[2]
Ideas
The project includes agent benchmarks and local tools for tracking cost, tokens, elapsed time and tests. The reported results describe specific tasks and agent configurations. They do not guarantee the same improvement in every project.[2]
Tools, writing & talks
Tools
Start here
Ponytailgithub.comInstructions, review commands and local session tools that encourage coding agents to reuse existing solutions and avoid unnecessary implementation.
Read the solution ladder and before-and-after examples first. Applying the instructions requires a supported coding agent and a repository whose conventions the agent can inspect.
Publishes repository[1][2]Original source