ayghri/i-have-adhd

If your coding agent regularly opens with "Great question!" and buries the fix three paragraphs down, this is a low-cost skill to try. It changes response shape, not the model's capabilities, so calibrate expectations before installing.

The problem

You ask a coding assistant where a token check goes wrong. It responds with a friendly preamble, a tour of the surrounding architecture, a tentative suggestion, a reminder to run tests, and a closing offer to dig deeper. The actual file and line number sit somewhere in the middle. You re-read, scroll, and extract the one actionable sentence yourself.

With this skill installed, the same exchange opens with the edit to make at a specific file and line, follows with numbered steps, and closes with a single concrete next move, such as pasting the first failing test line. The answer still contains technical reasoning, but the actionable part moves to the top.

What's different

This is not a plugin that changes how the agent works; it is a set of ten writing rules loaded as a skill. The rules target habits familiar to anyone who reads a lot of agent output: lead with the next action, number multi-step tasks, keep lists to five items, give minute-level time estimates, restate state each turn, and skip preamble, recap and closers. The full text lives in a single SKILL.md, so the whole mechanism is inspectable and editable.

The practical benefit is a cleaner transcript, which is most useful when you are scanning several replies a day rather than reading one carefully. The trade-off is durability. A commenter who tried the skill reported that the conciseness held for only a few turns before the model returned to verbosity, even with the skill named in a global instruction file and repeated during sessions (comment, 2026-09-08). Another commenter pointed out that Claude Code already ships a predefined Concise output style that leads with results, skips narrating, and keeps error-report content intact (reply, 2026-09-08). In other words, check whether your tool's built-in output style covers the need before adding another moving part.

The README is also candid that the rules are aimed at how an LLM should respond, not at clinical support: "No ADHD diagnosis needed." The name is a framing device for impatience with padding, not a claim of therapeutic benefit.

Who should try it

Good fit: you use a coding assistant interactively, you find yourself skimming or re-reading long replies, and you want a small, inspectable rule set you can tune by forking the skill file. Editing the rules is explicitly supported, which makes it reasonable for a team that wants a house style for agent output.

Poor fit: you already use your tool's concise output mode, or you need a guarantee rather than a tendency. The skill is instruction-level guidance to the model, and the reported drift after a few turns matters most for long sessions where consistency is exactly what you need. If your main annoyance is a specific tic, such as the model listing everything it did not do, a narrower rule may serve you better than a general bundle (comment, 2026-09-08).

Requirements and limitations

You need a coding assistant that supports skills or plugins. The README gives two paths: paste an install prompt into your CLI and let the agent follow the repository's instructions, or follow the linked installation instructions manually. The prompt-based path asks the agent to fetch and install code on your behalf; a commenter raised concern about that pattern and about installing from a lookalike repository (comment, 2026-09-08), which is a reasonable reason to prefer the manual path or to inspect the skill before installing.

Ongoing upkeep is minimal but not zero: customizing means forking, editing SKILL.md, uninstalling the upstream copy, adding your marketplace, reinstalling, restarting and re-invoking. The README includes those steps, and they are the part to budget for if you plan to tune it. There is no separate runtime dependency.

HotRepo's take

HotRepo first recorded this project on the daily board on 2026-07-21 at 6,279 stars; by 2026-10-27 it was at 58,972 with +182 that day. The last three recorded board days averaged +474/day against +2,407/day in the three before, so recorded momentum is fading. The project dates from 2026-05-13, so this is a young repository with a documented but short track record; persistence cannot be judged from these observations.

Before installing, read the ten rules in SKILL.md and check whether your assistant already has a concise output style. If it does, compare that first; the skill's main value is in being editable.