Strict, focused rules with the reasoning attached, so your skill files actually steer your agent instead of becoming documents it reads and then ignores.
How to package your taste into skill files that work. Encode process instead of output, pair rules with reasoning, be strict, and cut every line that doesn’t change behavior.
Run the installer in your terminal to add this skill to your project:
Omit the skill name to pick from all skills at once, and rename skills during install if you prefer different slash commands.
Once installed, use it when you're writing your own skill file and your agent keeps ignoring it. For example: