i-have-adhd Is a Claude Code Skill That Makes Your AI Stop Burying the Answer

Ask a coding agent to fix an auth bug and a common answer starts like this: "Great question! Let me think about this. Your auth flow has a few moving pieces..." three sentences later, you still don't know which file to open. i-have-adhd is a Claude Code and Codex skill built to cut that pattern out entirely: no throat-clearing, no "hope this helps," just the file, the line number, and the next step.

What actually changes

The skill rewrites how a coding agent structures its replies around 10 rules, the full text lives in its own SKILL.md: lead with the next action, number multi-step tasks, end with one concrete next step, suppress tangents, restate state every turn, give specific time estimates instead of "a bit," make wins visible, deliver errors matter-of-factly, cap lists at five items, and cut the preamble and recap entirely. The project's own before-and-after example makes the shift concrete: a paragraph explaining the auth flow, the token verification, and "you might also want to look at your dependency versions" becomes three numbered steps and a file path.

It installs as a real skill, not a system prompt hack. For Claude Code: claude plugin marketplace add ayghri/i-have-adhd then claude plugin install i-have-adhd@i-have-adhd, invoked with /i-have-adhd, or made to run on every session with touch ~/.claude/.i-have-adhd-always. Codex gets the equivalent through its own plugin marketplace command, invoked explicitly with $i-have-adhd or triggered implicitly when Codex judges a task benefits from it.

Why this is a real pattern, not a gimmick

The name is a deliberate framing choice, not a diagnosis requirement: the project credits The Adult ADHD Tool Kit by J. Russell Ramsay and Anthony L. Rostain as a loose basis, adapted for how an LLM should respond rather than how a person organizes a day. The actual mechanism it targets is one that shows up constantly in agentic coding regardless of who's reading the output: a model that explains its reasoning at length before committing to an action burns context and reader attention on narration instead of the thing that actually needs doing.

This connects directly to a benchmark AIOS Guide covered this same week: HANDBOOK.md, which found that AI agents given long, detailed standing instructions frequently fail to apply the right rule at the right moment, not because the rule wasn't there, but because a long document doesn't reliably change behavior. i-have-adhd is a small, sharply scoped version of the opposite bet: instead of a long standing document the agent might drift from, it's 10 rules, loaded as a skill, aimed at one specific failure mode (verbosity that buries the actionable part of the answer) rather than trying to govern everything at once.

Fork it before you trust it blindly

The skill is meant to be edited, not just installed as-is. The maintainer's own instructions are to fork the repo, edit skills/i-have-adhd/SKILL.md directly, then swap the fork in with claude plugin uninstall i-have-adhd, claude plugin marketplace remove i-have-adhd, and re-add your own fork under the same plugin name. That matters because "matter-of-fact errors" and "cap lists at 5 items" are opinionated defaults, not universal truths, a rule that helps one person's workflow can genuinely get in the way of another's, and the project makes tuning it a first-class path rather than an afterthought.

The honest caveat

There's no traction data or benchmark published for how much time this actually saves in practice, the evidence here is the design itself and the before/after example, not a measured study. If your agent's current output already reads the way you want, this is a skill to skip. If you've ever re-read three paragraphs of an agent's reasoning looking for the one line that told you what file to open, it's a direct fix for exactly that.

Free and open source under the MIT license, built by ayghri, 14,000-plus GitHub stars as of this writing, with translations into Chinese, Japanese, Korean, and Vietnamese.