TL;DR A great agent skill is not a pile of documentation. It is a tightly scoped SKILL.md with a description engineered for discovery, ruthless conciseness, anti-patterns stated up front, a checklist workflow, and a feedback loop. The format is an open standard that works across Claude Code, OpenAI Codex, Google Antigravity, Gemini CLI, and Cursor. This post synthesizes the official authoring guidance from Flutter, Anthropic, Google, and OpenAI into one recipe, hands you a complete copy-pasteable Flutter skill, and shows you how to actually evaluate it instead of guessing.
In my last article, I wrote about the official Dart and Flutter Agent Skills and why they stop your AI from writing 2022 Flutter. The most common reply I got was some version of the same question:
"Cool. How do I write my own?"
So I went and read the actual playbooks. Not the hot takes, the primary sources: Flutter's skill docs and eval framework, Anthropic's skill authoring best practices, Google's Antigravity skill docs, and OpenAI's Codex skill guide. The good news is they agree on almost everything. The better news is that the gap between a skill that works and a skill that gets silently ignored comes down to a handful of decisions, and most people get them wrong.
