i-have-adhd: How 143 Lines of Markdown and ADHD Neuroscience Fixed AI's Verbosity Problem (9,200+ GitHub Stars)

Forum topic · ✨步子哥 · 2026-08-03

Summary

i-have-adhd is a viral GitHub project that reached over 9,200 stars in two months using only 143 lines of Markdown and zero code. Created by an ML PhD, it works as a "skill" file that teaches AI coding assistants to stop burying actionable answers under filler like "Great question!" and "Hope this helps." Unlike earlier de-slop prompts, its ten rules are derived from five facts of ADHD cognitive neuroscience: limited working memory, the knowing-doing gap, difficulty with task initiation, uniform time perception, and dopamine scarcity. Rules include leading with the next action, numbering steps, giving concrete next actions, using minute-based time estimates, capping lists at five items, and banning preambles and closers. A pre-send checklist acts as a linter for AI output. The project reframes AI alignment as an output-layer intervention—reshaping response format rather than retraining models—and raises a real trade-off between action speed and user autonomy. Based on Ramsay & Rostain's CBT-based The Adult ADHD Tool Kit.

i-have-adhd: 143 Lines of Markdown, 9,200+ Stars, Zero Code

The Problem Every AI Coding User Knows

Ask Claude about a broken auth flow and you get a wall of text: greetings, restated questions, thinking-out-loud, hedged suggestions, and a cheerful sign-off. The actual instruction—"update the package and rewrite that function"—is buried in the seventh sentence.

The answer gets buried. That is what i-have-adhd, a GitHub project that went viral in July 2026, sets out to fix. An ML PhD built it from 143 lines of Markdown and 0 lines of code, earning 9,236 stars in two months. It is not a library—it is a "skill" file that teaches AI coding assistants how to respond.

After installing it, the same question gets:

> Run npm install jsonwebtoken@latest, then edit src/auth.ts:42. > 1. Open src/auth.ts > 2. Replace verifyToken (lines 42–58) with the snippet below > 3. Run npm test -- auth.spec.ts > > Next: paste the first failing line if any test fails.

First line = next action. Numbered steps. One concrete ending. No "Great question."

Why It Went Viral

9,236 stars in two months for 143 lines of Markdown is an abnormal ratio. stop-slop (a similar de-AI-flavor skill) took three months to reach 2,000 stars. The reason is not technical novelty—the pain point is universal. Every AI coding user has scrolled through filler to find one actionable line. Nobody built this because it sounded too simple. The author did—and, crucially, derived the rules from ADHD neuroscience rather than personal preference.

Five Brain Facts Behind the Rules

1. Working memory is small → Rule: *lead with the next action.* 2. Knowing the answer ≠ doing the answer → Rule: *number multi-step tasks.* 3. Initiation is the hardest step → Rule: *end with one concrete next step.* 4. Time estimates feel uniform → Rule: *use specific minute-based estimates.* 5. Dopamine is scarce → Rule: *make wins visible.*

The README notes it is "loosely based on *The Adult ADHD Tool Kit* by J. Russell Ramsay and Anthony L. Rostain," a CBT-based adult ADHD guide.

The Ten Rules

1. Lead with the next action 2. Number multi-step tasks 3. End with one concrete next step 4. Suppress tangents 5. Restate state every turn 6. Specific time estimates 7. Make wins visible 8. Matter-of-fact errors (no drama) 9. Cap lists at 5 items 10. No preamble, no recap, no closers

Designing for the most vulnerable users often benefits everyone—the accidental Universal Design principle.

The Pre-Send Check

The skill's best part is a four-deletion, one-verification checklist:

Delete: the first sentence if it announces what you're about to do; the last sentence if it asks "anything else?"; any "by the way" asides; vague adverbs adding no information ("maybe," "possibly").

Keep: genuine uncertainty—removing it manufactures false confidence.

Replace: idioms and metaphors with literal actions.

Verify: if a reader reads only the first and last lines, do they know (a) what to do next and (b) what just happened? If yes, send.

It is a linter for AI output—like code linters, you don't memorize rules; you run the check.

CBT for LLMs

CBT's core logic: don't change ability—change the environment to compensate. i-have-adhd applies this in reverse: instead of changing the ADHD user's environment, it changes the AI's output environment to match ADHD cognition. Not "make it shorter" but "make the output shape match how the brain processes it." Shape matching, not brevity.

The Criticism: Cost of Deleted Context

The sharpest critique: removing the "why" strips the context users need to judge correctness, learn the domain, and catch AI errors. It's a real trade-off—action speed vs. autonomous judgment. For ADHD users, whose working memory is precisely the bottleneck, the bet may be right. For neurotypical users, it risks turning them into executors of AI suggestions rather than collaborators. A five-dimension scoring framework (correctness 35%, autonomy 25%, actionability 20%, plus two more) captures this: "good output" is not one-dimensional.

Connection to Sycophancy Research: Output-Layer Alignment

Compared with other interventions:

| Intervention layer | Example | Cost | Effect | |---|---|---|---| | Training | RLHF reward adjustment | Very high (retraining) | Global | | Inference | Prompt frameworks | Medium | Scenario-specific | | Output | i-have-adhd skill | Very low (143 lines of Markdown) | Output shape |

i-have-adhd proves output-layer intervention is extremely cost-effective—reshaping the output of a model without retraining. Community-driven alignment can work bottom-up.

The Deeper Insight: The Knowing-Doing Gap

Fact two is the most profound: *knowing the answer is not doing the answer.* This is not just an ADHD problem—it is a universal human one. The width of the gap between knowing and doing is set by environment design. If the answer is in line one, you act; if it's in sentence seven, maybe you don't. The shape of AI output determines whether the user can cross the knowing-doing gap.

FAQ

Who is this for? Practitioners, researchers, and students interested in AI, machine learning, and developer tooling.

Key takeaways? AI verbosity buries actionable answers; i-have-adhd derives ten output rules from ADHD neuroscience; a pre-send checklist works like a linter; output-layer alignment is extremely cheap; there is a real trade-off between action speed and autonomy.

Open source? Yes—MIT licensed, links above.