A skill for stopping AI agents from burying the answer

ayghri/i-have-adhd hit Hacker News at 489 points, and it has 33,100 stars and 2,000 forks. The whole thing is ten prompt-level rules for AI coding agents: lead with the next action, number multi-step tasks, cut tangents, kill preamble.

The stakes are about attention, not accuracy. An agent that buries "run this command" under three paragraphs of context costs a developer the same scan-and-search tax every single reply. The repo rewrites that output into direct instructions, something like "Run npm install jsonwebtoken@latest, then edit src/auth.ts:42" instead of a wall of explanation before the command shows up.

What makes it different from a personal CLAUDE.md tweak is portability. It installs across Claude's plugin system, Cursor skills, Gemini, Qwen, and Kimi. Most agent-behavior fixes get written once for whatever tool someone happens to use and never travel. This one was built to run the same rules on five harnesses that don't share a config format or a prompt structure.

That portability is also the bet worth watching. Five different agent platforms don't expose the same hooks for intercepting and rewriting output, so a rule set that behaves identically on all of them either found a common seam across all five, or one of the five breaks the pattern the next time its model changes how it formats replies. The repo doesn't say which.