I Have ADHD: Skill That Stops Claude from Burying the Answer

You ask your coding agent why the auth flow is broken. It opens with "Great question!", walks you through the moving pieces, mentions the function that looks suspicious, suggests an approach, adds a thought about your dependency versions generally, and signs off with "Hope this helps!"

Somewhere in there was the fix. You scrolled past it twice.

i-have-adhd is an open-source skill by ayghri that stops this. It does not make the agent smarter or change what it knows — it changes the shape of what comes back. The install command first, the file path second, numbered steps, one next action, nothing else. The repo tagline is worth repeating: ADHD-friendly outputs, no ADHD diagnosis needed. It is MIT licensed and sits at over 21,000 stars.

What actually changes

The repo's own before-and-after is the clearest pitch. The before is a paragraph of throat-clearing that mentions the broken function somewhere in the middle and ends with an offer to dig deeper. The after opens with the install command and the exact file and line to edit, gives three numbered steps, and closes with one instruction: paste the first failing line if a test breaks.

Same information. The second version you can act on without reading it twice.

The ten rules, grouped

The skill is ten rules in a SKILL.md file. They fall into three jobs.

Shape the answer. Lead with the answer or next action — command, path or snippet on line one. Number multi-step work, one bounded action per step. End with a single next action that takes under two minutes.

Cut the noise. No preamble, no recaps, no closers. Cap lists at five items. Finish the current issue before raising a new one — which kills the tangent where the agent fixes your bug and then volunteers opinions about your folder structure.

Keep you oriented. Restate progress every turn, in the style of step three of five done. Give time estimates in concrete units rather than "a bit". After a change, show what now works. Report errors as location, cause and fix, without drama.

The full wording lives in the repo's SKILL.md, and it is short enough to read in a minute before you install anything.

The exceptions are the good part

A terseness rule applied without judgement is worse than verbosity. This one carries its own escape hatches: it explains fully when you actually ask for an explanation, confirms before destructive actions, and asks one short question when a request is ambiguous instead of guessing.

The best rule in the set: after three failed fixes, stop and name the assumption you are doubting. That is a direct hit on the worst agent failure mode — the confident fourth, fifth and sixth attempt at the same wrong theory.

Installing it

It ships for a long list of harnesses — Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot, Zed, OpenCode, Qwen Code, Kimi, Hermes, Pi, Antigravity. The three most likely to matter:

Claude Code

claude plugin marketplace add ayghri/i-have-adhd

claude plugin install i-have-adhd@i-have-adhd

Then type /i-have-adhd. If it does not show in autocomplete, restart — the plugin index is read at startup.

Codex

codex plugin marketplace add ayghri/i-have-adhd --ref main

codex plugin add i-have-adhd@i-have-adhd

Invoke it with $i-have-adhd rather than a slash. Codex will not activate it on its own.

Cursor, OpenCode and anything else that reads agent skills

npx skills add ayghri/i-have-adhd

Add -g for every project, or -a cursor to target one agent. Start a new chat afterwards, since skills are indexed at session start. Gemini CLI is the odd one out — no marketplace, so you either curl a TOML file into ~/.gemini/commands for the opt-in version, or install it as an extension, which is always-on from message one.

On-demand by default — and how to make it permanent

This is the design decision most people miss. In Claude Code, Codex and Qwen Code, installing the plugin changes nothing at all. The skill sets disable-model-invocation, so the agent cannot pull it in on its own — if you did not turn it on, it is off. Some other harnesses load every skill description at startup and may activate it themselves.

Once invoked, the rules hold for that session. Saying stop adhd mode or normal mode turns them off again.

For always-on in Claude Code, create a flag file and a SessionStart hook loads the full ruleset from message one of every session:

touch ~/.claude/.i-have-adhd-always

Delete the file to go back to on-demand. Everywhere else, the repo gives you a ten-line block to paste into your persistent rules file — AGENTS.md for Codex and OpenCode, GEMINI.md for Gemini, copilot-instructions.md for Copilot, User Rules for Cursor. Same rules, no plugin machinery.

Fork it and change the rules

This is where it gets interesting for anyone who is not writing code. The rules are plain English in one markdown file. Fork the repo, edit skills/i-have-adhd/SKILL.md, then remove the upstream copy and add your own — the fork and the original share a name, so you have to drop one before installing the other.

For a media team, the same skeleton rewrites cleanly: lead with the campaign object or metric rather than the command, cap the list at five recommendations, restate which of the five accounts you have worked through, name the one change to make next. The structure is the reusable part, not the code-specific wording.

What it does not do

It shapes output. That is the whole mechanism, and worth being clear about before you install it expecting more.

• Accuracy is untouched. A wrong answer delivered as a numbered list is still wrong, just faster to reach and easier to act on before you have noticed.
• Rules drift over long sessions. The repo's own troubleshooting says as much: open a new session, and if it still preambles, tighten the wording in the SKILL.md yourself.
• Terse is wrong for some work. Anything where the reasoning is the deliverable — a design decision, a tradeoff, an explanation you will pass to someone else — wants the long version. Ask for it explicitly and the skill steps aside.

On the name

The repo is upfront that no diagnosis is required to want this, and credits The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain as loose inspiration — adapted for how a model should respond rather than how a person should organise their day. It is a framing borrowed from work on executive function, not a medical tool, and it does not claim to be one.

Which probably explains the 21,000 stars. Almost nobody enjoys reading "Great question!" before finding out which line to edit.

Links

• i-have-adhd on GitHub
• Install instructions for every supported agent
• The full SKILL.md ruleset