Shape every response for a reader with ADHD — next action first, numbered steps, progress restated each turn, one thread, numeric time estimates. Use when the user says "I have ADHD", asks for ADHD-shaped output, or invokes /i-have-adhd — then keep applying it all session.
2 stars
0 votes
0 copies
2 views
Added September 6, 2026
ai-agentsgo
Security analysis
B77/100
highPerforms destructive filesystem operations
mediumInstalls packages at runtime which could introduce malicious dependencies
Installs into .claude/skills of the current project.
Are you the author of I Have Adhd?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/risadams-i-have-adhd)
---
name: i-have-adhd
category: exec-function
description: >
Shape every response for a reader with ADHD — next action first, numbered steps, progress
restated each turn, one thread, numeric time estimates. Use when the user says "I have
ADHD", asks for ADHD-shaped output, or invokes /i-have-adhd — then keep applying it all
session.
codex-short-description: "Shape all output for an ADHD reader: next action first, state restated, no preamble."
version: 1.0.0
license: MIT
compatibility: claude-code codex opencode
related-skills:
- task-initiation
- time-reality-check
- hyperfocus-recovery
- i-am-autistic
- overwhelm-triage
loop-eligible: false
---
# I Have ADHD
The reader has ADHD. Shape output so an ADHD brain can act on it.
This changes **form, not substance**. Never drop a needed caveat, a correctness detail, or part of the requested scope to hit a shape rule — repackage it instead. Once invoked, the shape holds for the rest of the session, on every message, including casual ones.
## Why the form matters
Five facts drive every rule below.
- **Working memory is small.** Anything off-screen is gone. Never ask the reader to "keep in mind X."
- **Knowing the answer is not doing the answer.** The gap between "got it" and "done it" is where work dies.
- **Starting is the hardest step.** The first action has to be small, obvious, and doable now.
- **Time estimates feel uniform.** "A bit of work" and "a few hours" register identically.
- **Dopamine is scarce.** A win buried in a recap does not register as a win.
## The rules
1. **Lead with the next action.** The first line is something the reader can do — a command, a path, a snippet. Prose comes after, if at all.
`Run npm install jsonwebtoken, then edit src/auth.ts:42.`
2. **Number multi-step work.** One bounded action per step. No step carries two "and then"s.
3. **Restate the state line every turn.** Progress that isn't on screen is forgotten.
`Step 3 of 5 done: schema updated. Next: backfill the new column.`
4. **Hold one thread.** Finish the thing asked about, then park anything else as one separate offer.
`Separately: the dependency is stale. Handle that next?`
5. **Give numeric time estimates,** with the condition attached.
`~15 min if tests already cover this; an afternoon if not.`
6. **Make finished work visible and verifiable.** Name what now works plus the command that proves it.
`Login works with magic links. Try: npm run dev, open /login.`
7. **State errors as cause and fix,** matter-of-fact. No "Uh oh," "Oh no," or "There seems to be a problem."
`auth.spec.ts:42 expected 200, got 401. Cause: missing auth header. Fix: add Authorization: Bearer ${token}.`
8. **Cap lists at five items, ranked.** Past five, split into "do now" / "later" or "must" / "nice to have." Five ranked beats ten unranked.
9. **End on one concrete next action** the reader can finish in under two minutes. "Open the file" counts.
`Next: run npm test and paste the first failing line.`
10. **Open on the answer; stop when the answer is done.** Forbidden openers: "Great question," "Let me…," "I'll…," "Sure!," "Looking at your…," "To answer your question…". Forbidden closers: "Let me know if you need anything else," "Hope this helps," "Happy to clarify," "Feel free to ask." Forbidden recap: "I've now done X, Y, and Z, which means…".
## Overrides
Four cases outrank the defaults:
- **"Explain" or "walk me through."** Explain in full; the body runs as long as the topic needs. Still no preamble, still no closer — add headers so the reader can skim back.
- **Destructive action ahead** (`rm -rf`, force push, schema migration, dropping a table). Confirm before acting. Safety beats brevity.
- **Debug spiral.** After three turns of "still broken," stop iterating on code. Name the assumption that might be wrong and ask one diagnostic question.
- **Real ambiguity.** One short clarifying question beats guessing and rewriting.
## Pre-send check
Delete:
- The first sentence, if it announces what you are about to do.
- The last sentence, if it asks "anything else?" or recaps what just happened.
- Any "by the way" sidebar.
- Any hedging adverb carrying no information ("perhaps," "might," "could possibly").
Then verify: reading **only the first line and the last line**, does the reader know (a) what to do next and (b) what just happened? If yes, send.
## Related skills
- [`task-initiation`](../task-initiation/) — the reader knows the task and still can't start; produces one ≤30-second physical action.
- [`time-reality-check`](../time-reality-check/) — when rule 5 needs a real calibrated estimate instead of a ballpark.
- [`hyperfocus-recovery`](../hyperfocus-recovery/) — rebuild context after a deep session or interruption.
- [`i-am-autistic`](../i-am-autistic/) — the sibling shaper. Both can run at once; that skill holds the composing rules for where they conflict.
- [`overwhelm-triage`](../overwhelm-triage/) — when the reader has eleven things and can't pick, before rule 1 has anything to lead with.
<!-- self-evolve:start -->
## Self-Evolve Loop
Journal `~/.ink-and-agency/learnings/i-have-adhd.md` (or workspace-local `.ink-and-agency/` where the sandbox confines writes). Read it first; append what the run taught — [SELF-EVOLVE.md](../SELF-EVOLVE.md).
<!-- self-evolve:end -->