Skill review · by Ayoub Ghriss (GitHub: ayghri), ML researcher, University of Colorado Boulder
I Have ADHD Claude Skill
The "i have adhd" skill for Claude is ayghri/i-have-adhd, an open-source ten-rule ruleset that reshapes how a coding agent writes its answers, published by Ayoub Ghriss, a machine-learning researcher at the University of Colorado Boulder. It stands at 26,301 GitHub stars as of September 1, 2026, is MIT-licensed, and was last pushed August 26, 2026. It is not a clinical tool and does not adapt to an individual diagnosis; it formats every response so an ADHD reader can act on it: the next action first, numbered steps, restated progress each turn, concrete time estimates, no preamble, no closing pleasantries. The one thing to know before installing: it ships off by default. The skill sets disable-model-invocation, so nothing changes until you type /i-have-adhd in a session — or create the ~/.claude/.i-have-adhd-always flag file, which loads the rules at every session start.
Founder, Automation Squad ·
Install
- 1In a terminal, run: claude plugin marketplace add ayghri/i-have-adhd
- 2Then run: claude plugin install i-have-adhd@i-have-adhd
- 3In a Claude Code session, type /i-have-adhd to switch it on; saying "stop adhd mode" switches it off for that session
- 4Optional always-on: run touch ~/.claude/.i-have-adhd-always — a SessionStart hook then loads the ruleset in every session; rm the file to go back to on-demand
- 5Verify with: claude plugin list
What it does
- Action-first answers
- The first line of every response is something you can do — a command, file path, or code snippet — with prose after it, if at all (SKILL.md rule 1).
- Bounded structure
- Multi-step work becomes a numbered list with one action per step, lists are capped at 5 items, and every open-ended reply ends with one next action doable in under two minutes.
- Working-memory support
- Progress is restated every turn ("step 3 of 5 done"), completed work is shown concretely rather than buried in a recap, and time estimates come in specific units instead of "a bit of work".
- Noise removal
- Preambles, recaps, closing pleasantries, and mid-task tangents are banned; errors are reported flatly as location, cause, and fix.
- Escape hatches
- The rules yield when they should: full explanations when you ask to be walked through something, confirmation before destructive actions, a circuit-breaker after three failed fixes in a debug spiral, and one clarifying question on genuinely ambiguous requests.
Common questions
- Is i-have-adhd an official Anthropic skill?
- No. It is an MIT-licensed community project by Ayoub Ghriss, a machine-learning researcher at the University of Colorado Boulder, hosted on his personal GitHub account. Anthropic has no involvement.
- Is this the ADHD skill that supposedly makes Claude "think 2x better"?
- No. That claim belongs to UditAkhourii/adhd, an unrelated tree-of-thought reasoning skill covered by The New Stack, where outside experts questioned the benchmark. i-have-adhd changes how answers are written, not how the model reasons, and publishes no performance claims.
- Does it work automatically after installing?
- No. The skill ships with disable-model-invocation set, so you must type /i-have-adhd in each session to activate it. Creating the ~/.claude/.i-have-adhd-always flag file makes a SessionStart hook load it every session, though a known issue reports that hook silently failing in the macOS desktop app.
- Do I need Claude Code, or does it work with other coding agents?
- The same SKILL.md ships with native install paths for nine agents: Claude Code, Codex, Gemini CLI, GitHub Copilot, Kimi Code, OpenCode, Hermes, Cursor, and Antigravity. There is no documented install for the Claude.ai web or desktop chat apps.