The Claude Code ADHD Skills I Actually Run (Install in 2 Minutes)

If you searched for a Claude ADHD skill and want a file on disk in the next ten minutes, this page is the file. The repository is public, MIT licensed, has no installer and no dependencies, and the three SKILL.md files in it are byte-for-byte copies of the ones in my own ~/.claude/skills/ directory. Nothing in it is a demo written for a blog post.

The two-minute install

There are three ways in. Pick the one that matches how you already work.

Let the agent do it. Paste this line into Claude Code, or into any coding agent that can read a repository:

Install the adhd-skills plugin from https://github.com/Citability/claude-code-adhd-skills, refer to the repo's AGENTS.md for instructions.

The repository ships an AGENTS.md that tells the agent where the skills live, which install path to use, what it must not touch, and how to verify the result.

Plugin marketplace (Claude Code 2.x):

claude plugin marketplace add Citability/claude-code-adhd-skills
claude plugin install adhd-skills@adhd-skills

Plain copy. Claude Code loads any directory under ~/.claude/skills/ that contains a SKILL.md, so the install is a clone and a copy. This works on any version.

git clone https://github.com/Citability/claude-code-adhd-skills.git
mkdir -p ~/.claude/skills
cp -r claude-code-adhd-skills/mirror claude-code-adhd-skills/librarian claude-code-adhd-skills/standing ~/.claude/skills/

Whichever path you took, restart Claude Code, then type:

/mirror

Repository: github.com/Citability/claude-code-adhd-skills

To install for one project only, copy the three folders into that repo’s .claude/skills/ instead of your home directory. Claude Code reads both locations.

I send a short note when a skill in the set is renamed, retired, or added. It is a plain newsletter, and nothing on this page depends on it.

Which ADHD skill is which

Three unrelated things are currently called “the Claude ADHD skill”, and nothing on the search results page tells them apart. The AI Overview describes one of them, the second organic result is a different one, and the searcher can end up installing a skill that does something other than what they were told.

	i-have-adhd	adhd	This repo
Author	Ayoub G.	UditAkhourii	Chudi Nnorukam
What it changes	The output format	The reasoning process	Session review, weekly standing, and codebase navigation
Relation to the condition	Built for readers with ADHD	A metaphor for divergent thinking	Built for a specific executive-function gap each
Published benchmark	None	Six problems, self-judged	None. First-hand daily use instead.
Runs on	Every reply	Open design problems	End of session; end of week; abstract requests in a codebase

i-have-adhd is ten writing rules: lead with the next action, number the steps, suppress tangents, restate state, make wins visible, cap lists at five, no preamble. It changes where the answer sits in the reply and nothing else. If Claude’s answer being in paragraph four is the same to you as the answer not being there, that skill fixes it in one step. It publishes no benchmark and does not need one, because you can feel a formatting rule working inside a day.

adhd replaces a single chain of reasoning with five independent branches under different frames, scores the ideas, discards the ones it flags as traps, and develops the survivors. The name is a metaphor. It has nothing to do with the condition and does not claim to. Its published benchmark shows large gains on breadth, novelty and trap detection, which are close to asking whether the method ran, and a gain of under one point out of ten on builder usefulness. It lost the one problem that already had a known answer, a retry and timeout strategy, because a wider search walked it away from the standard solution. Use it on open problems. Do not use it on problems with a known answer.

This repo is neither. It is three skills that each patch one named executive-function deficit, described below, and it exists because those three gaps cost me more than output formatting ever did.

To check which one you already have, run /plugin in Claude Code and read the names. Shorter replies that open with the next step mean the formatting skill. Long option lists with a section headed traps mean the reasoning skill. A three-line readback that starts with Hearing: means you have librarian from this repo. A weekly report where every number carries the command and the date range that produced it means standing.

mirror: the counter to discounting the positive

The gap it patches: the cognitive distortion of discounting the positive.

“Discounting the positive” is a cognitive-behavioural-therapy term for deleting your own wins from the ledger. You shipped three features and your brain renders only the bug. ADHD bundles this so reliably that it feels like realism. It is a measurement error, and it starves a reward system that is already under-fuelled.

mirror pulls the actual record before it says anything: the git log since the session started, the diff of uncommitted work, the decisions made in the conversation, the dead ends ruled out. Then it reflects that back, specific and unembellished. Not “you did great”. “You shipped three commits, fixed the auth redirect, and ruled out the caching theory. That is the bug closed.” It names invisible work explicitly, because an hour spent understanding a race condition is the reason the fix took ten minutes.

When it fires. “I got nothing done.” “Today was a waste.” “I am so behind.” “Was this session even worth it?” Any end-of-session reflection where the tone is self-erasing. It stays quiet when you want a genuine critique of the work, and it does not pathologise a good mood.

What it refuses to do. Flatter, argue with the feeling, invent accomplishments, or turn into a pep talk. If little genuinely got done, it says so and locates the cause accurately, which is usually the plan and not the person.

Try it now, at the end of a session:

I feel like I got nothing done today. How did it actually go?

librarian: the map you do not have to hold

The gap it patches: working-memory overload from navigating a codebase.

Working in a large codebase means holding a map of it in your head: where things live, what calls what, which file owns which concern. ADHD working memory does not hold that map, so every navigation is a fresh expensive search, and the expense becomes a reason to avoid unfamiliar parts of the code.

librarian holds the map instead. You describe the goal in your own words, “why is startup slow”, and it walks the project’s knowledge graph, checks each node it touches for drift against the current code, and answers in exactly three lines:

Hearing:  <the goal, in your words>
Impacts:  [the knowledge nodes involved]
Proposal: <one concrete change>  [Verified | Inferred | Target]

Then, if the change is reversible and every node it relied on was fresh, it makes the change. If anything drifted or the change is not reversible, it stops after the readback and asks. After acting, it writes what it learned back into the graph, so the next walk starts further along.

When it fires. The request names a goal or a behaviour and does not name a file. “Make the import faster” fires it. “Edit server.ts:120” does not, and it says so and steps aside.

The dependency, stated plainly. librarian reads a knowledge graph at docs/codex/ in your repo: an index file, one markdown node per concept, and an optional principles file. If your repo has no codex, the skill has nothing to walk. It will tell you so and name /codex-bootstrap, which is not in this repo. Install mirror first. Install librarian when you have a codex, or when you are ready to build one.

standing: the week against your own month

The gap it patches: comparing yourself to an imagined peer who ships every day.

mirror answers “did I do anything today”. standing answers the slower, heavier question, “am I getting anywhere”, and it refuses to answer it with a feeling. The row from the repository README, quoted verbatim:

Skill	The gap it closes	Standalone?
standing	Comparing yourself to an imagined peer who ships every day. Reads your session transcripts and git, counts the invisible work, and places the week against your own trailing month, with instrument and window on every number.	Yes. Works from git and Claude Code transcripts. Field comparison needs a cited docs/standing/benchmark.md.

It reads the last few Claude Code transcripts for the current directory plus git log, sorts the week into four buckets (shipped, invisible, open, abandoned), and prints the week beside your own trailing four weeks, cut on the same weekday so the medians compare. Every number in the report names the command and the full date range that produced it, and every derived number shows its arithmetic. If you have not written a benchmark file with a source for your field, it says Unknown rather than quoting a figure from memory.

When it fires. “How am I doing this week?” “Am I behind?” “Is this going anywhere?” Any weekly or monthly review where the comparison class is imaginary.

What it refuses to do. Invent a peer, quote an industry number without a citation, or restate a transcript’s number as if it had re-measured it.

The README’s Before/After table for this skill is two real runs on the same repository and the same day. The stock reply counted 92 commits and never said which window or command; the skill counted 85 with the git log call and full timestamps printed beside it, and all three headline numbers matched an independent recount. The skill’s first draft failed its own kill criterion on nine numbers before that discipline was added. The second run is the one in the README.

Try it now, on a Friday:

/standing

What connects them

All three skills read the same underlying record. mirror and standing read git, and standing adds the session transcripts Claude Code already keeps under ~/.claude/projects/, so the work that left no diff still counts. librarian walks the per-project knowledge graph and grows it, and mirror uses that graph as the benchmark for progress when one exists. The longer explanation of why that compounds is in the skills-for-ADHD post, and the configuration layer underneath both is in the CLAUDE.md guide for ADHD developers.

Install troubleshooting

• The slash command does not appear. Restart Claude Code. Skills are read at startup. Then confirm ~/.claude/skills/mirror/SKILL.md exists as a real file. A broken symlink looks like a directory in ls and loads nothing.
• It fires when you did not want it. The description: line at the top of each SKILL.md is what Claude reads to decide whether to pick the skill up on its own. Narrow that sentence. You can always invoke by name regardless.
• librarian says there is no codex. Expected on any repo without docs/codex/. That is a dependency, not a bug. Use mirror in the meantime.
• standing says the field comparison is Unknown. Expected until you write docs/standing/benchmark.md with a cited source. The skill still measures the week against your own trailing month.
• You want it in one project, not globally. Copy the folders into that repo’s .claude/skills/ and commit them. Everyone who clones the repo gets the skills.
• You edited a skill and want to share the edit. Fork the repository. It is MIT. I only merge changes to skills I run, so a pull request that changes what a skill does will get a slower answer than one that fixes how it reads.

Why only three

An earlier post on this site described five ADHD skills. Two of those five, mirror and librarian, are in this repo, and standing joined them on 2026-09-04. The other three are not, because I do not run them any more: one was retired, one never got past its design document, and one turned out to be a Claude Code built-in rather than a skill of mine. The rule for this repository is that a skill is listed only while it is present on my disk and in use, and it changes the same day a skill is renamed or retired. That is a smaller set than a listicle wants. It is also the only set I can vouch for.