Why it mattersA team using coding agents daily loses hours to preamble and closers in the reply, and moving the action to the first line is the smallest workflow change that pays back every turn.
The ayghri/i-have-adhd repository, an agent skill that rewrites a coding assistant's output so the reply leads with the next action, reached the front page of Hacker News this morning with 67 points and 29,156 stars on GitHub. The project has been in development since 13 May 2026, and a Windows import fix landed at 17:08 UTC today, so it is actively maintained.
The ten rules
The skill is a single SKILL.md file that adds ten output rules to the agent's system prompt. The full list, from the repository README: lead with the next action, number multi-step tasks, end with one concrete next step, suppress tangents, restate state every turn, give time estimates in concrete units, make wins visible, keep errors matter-of-fact, cap lists at five items, and no preamble, no recap, no closers.
The README ships a before-and-after that reads well. The "before" is 89 words that open with "Great question! Let me think about this" and end with "Hope this helps! Let me know if you want to dig deeper". The "after" is 40 words: the command to run, three numbered steps, and a next action if a test fails. It is the same information, shaped to be acted on rather than read.
Thirteen supported harnesses
INSTALL.md lists install commands for Antigravity, Claude Code, Codex, Gemini CLI, GitHub Copilot for VS Code and the Copilot CLI, Hermes, Kimi Code CLI, OpenCode, Pi, Oh My Pi, Qwen Code, and Zed. Cursor, Amp and any other agent-skills harness get a generic entry.
For Gemini CLI the skill can also be pinned as always-on by pasting the ten rules into ~/.gemini/GEMINI.md. Claude Code users invoke it with /i-have-adhd. The repository is MIT-licensed, README is translated into seven languages, and 1,808 forks are recorded against the main branch.
What it is loosely based on
The credits line names The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain, adapted for how a language model should respond. The name is deliberately loud, and the rules themselves are a general output discipline that any team can turn on for any user.
The traction is worth reading against the release cadence. 29,156 stars in 118 days is a real number produced by installs, and the 194 commits on the main branch confirm the project is still being maintained.
The reason a team already running Claude Code, Cursor or Codex should read this is that the ten rules cost nothing to try. A fork that keeps three of them and drops seven is still a workable answer. The hardest one to keep is the closer rule, because a plain "done" reads as rude to a human and reads correctly to a developer scanning ten replies a day.
Source
ayghri/i-have-adhd on GitHub. Hacker News discussion, 8 September 2026.
Source: GitHub
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually shipping with them. Short, and only when there is something worth reading.
