ArticleSeptember 3, 2026

i-have-adhd: skill so that the agent responds with the first line

Ten rules in one SKILL.md: action from above, no introductions, the list is no longer than five points. The form of the answer is corrected, the correctness is not, and it is better to throw out the requirement to name the deadlines right away.

Trending GitHub — repository ayghri/i-have-adhd. A skill for code assistants that changes the form of the answer: first line action, numbering for multi-step problems, no introductions or conclusions, list no longer than five items, specific time estimates. The rules are in one file skills/i-have-adhd/SKILL.md, next to it are layouts for different environments - .claude-plugin, .cursor/skills, .agents/plugins —and catalogs tests i evals. Installation is described in INSTALL.md, MIT license. Example from the README: instead of discussing the nature of the authorization problem, “do npm install jsonwebtoken@latest, then edit src/auth.ts:42».

What is being repaired

Form. If your agent responds with three screens of which two lines are needed, the skill will remove the binding. Savings are considered simple: open the last ten responses from the agent and note on which line the useful information begins. Consistently below fifth - place it.

There is also something that most other people's skills do not have: the evals catalogue. A style rule can be measured to see whether it worked or not. This is the only way to distinguish a skill that actually changes the output from a set of good wishes in a markdown. If you write your skills, copy this habit first, not the text of the rules.

Where it breaks

The “specific time estimates” rule is the most dangerous in the set. The model does not know how long it will take you to edit, and in response to a direct request to name the deadline, it will come up with one. Get a cheerful “15 minutes” for a task where half a day will be spent on one playback. This is exactly the mode in which the assistant sounds confident and makes mistakes.

The five-point limit cuts down long procedures. Migration by eight steps does not become five steps just because the style dictates so. It becomes a five-step answer, and you will find the missing three steps in the product.

The ban on introductions and conclusions also introduces reservations. The phrase “will work if you have X enabled” lives exactly in the paragraph that the skill tells you to delete. For someone new to the project, this is more expensive than the lines saved.

Practical conclusion: fork, don't bet as is. Remove the requirement to estimate deadlines. Replace the five-point limit with the formula “exactly as many steps as needed, zero unnecessary words.” Add your own rule: if you’re not sure, say it in one line, and don’t smooth out the wording.

Self SKILL.md is a good template at the same time. One file, a dozen rules, a couple of examples of “what was and what is now.” This is exactly what you need to write a skill that matches the way your team speaks: your own terms, your own ticket format, your own requirements for the commit message. Start with one rule, run it on ten problems, leave it if you see a difference.

Rule: the skill corrects the form of the answer and does not affect its correctness - from someone else’s style, first of all, throw out the point that forces the model to name numbers.

  • agents
  • claude-code
  • skills
  • prompting