UditAkhourii

UditAkhourii / adhd

Public

ADHD — a skill for coding agents. Tree-of-thought with pruning, built on the Claude Agent SDK. Fans out parallel divergent thoughts under different cognitive frames, scores, prunes traps, deepens the survivors. The no-brainer skill for creative and interdisciplinary work.

uditakhourii.github.ioadhd adhd agents ai ai-agents brainstorm
16
0
89% credibility
Found May 26, 2026 at 16 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

ADHD is a skill that makes AI coding assistants explore problems from many different angles before deciding on a solution. It works by running several parallel 'thought branches' where each branch thinks about the same problem from a completely different perspective — like a hardware engineer, a biologist, a skeptic, or someone constrained by zero budget. These branches don't see each other during this phase, which prevents the usual problem where one idea anchors all the others. A separate scoring pass then ranks all the ideas, marks tempting-but-dangerous traps, and deepens the most promising ones into full sketches with risks and first steps. It ships as a drop-in skill for AI coding tools, a command-line tool, or a library for developers building their own AI agents.

How It Works

1
💬 You hit a creative wall

You're working on a design decision or fuzzy problem with your coding assistant, and it keeps giving you the same obvious, forgettable answer.

2
🔍 You hear about a better way to think

A colleague mentions ADHD — a way to make AI agents explore many different angles before settling on a solution, like having a brainstorm room instead of just one person.

3
⚡ You add it to your AI assistant in one click

You drop a special skill file into your assistant's settings. Nothing to install, no passwords needed — it just starts working alongside your normal workflow.

4
🌐 Your assistant branches out in parallel

When you describe your problem, your assistant suddenly thinks like a hardware engineer, a regulator, a 10-year-old, and a biologist all at once — in separate parallel thoughts it never shares between them.

5
Different minds see the same problem differently
🔧
The engineer sees constraints

What latency, memory, and physical limits matter here?

🔬
The biologist borrows from nature

How would an immune system, ant colony, or neural network solve this?

😈
The adversary looks for traps

What could go wrong, and how would someone break this on purpose?

💰
The budget-mind goes extreme

What would work with zero money? What would work with infinite resources?

6
🏆 A separate critic scores and prunes

After all branches finish, a fresh perspective ranks every idea by how new it is, how practical it would be, and how well it fits your problem — and marks any that look tempting but are actually traps.

✨ You get the unexpected-but-workable ideas

The top survivors get expanded into full sketches with risks and next steps. You see clusters of related approaches, the non-obvious pick flagged for attention, and a provocation that challenges your assumptions — all things a single-pass assistant would never surface.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 16 to 16 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is adhd?

adhd is a skill for coding agents that forces them to think wide before they think deep. Instead of letting an agent latch onto the first reasonable answer, it fans out parallel branches under deliberately different cognitive frames--hardware engineer, biologist, speedrunner, 3am on-call engineer, and others--then scores, clusters, and deepens only the survivors. Built in TypeScript on the Claude Agent SDK, it ships three ways: as a drop-in agent skill, a Node library, and a CLI tool. You can call it with `adhd "design a rate limiter"` from the terminal or integrate it into your own agent loop.

Why is it gaining traction?

The hook is the generator-critic wall: most prompting techniques mix generation and evaluation in the same pass, which lets the inner critic strangle creative ideas before they fully form. adhd enforces a hard separation--branches never see each other during divergence, eliminating anchoring by construction. The 15 cognitive frames are the real differentiator; they don't just vary the next step, they re-ask the entire problem from a fundamentally different vantage point. The eval suite that ships with it is also notable--it benchmarks adhd against a single-shot baseline using an LLM-as-judge, which gives the approach some empirical teeth.

Who should use this?

Senior engineers and architects facing open-ended design decisions: sharding strategies, API surface design, naming, fuzzy debugging, or anything where you'd say "give me a few ways to..." The sweet spot is problems where the textbook answer is probably a trap. Not useful for bug fixes with known root causes or anything a junior would Google in two minutes. If you're building complex multi-agent systems, the library API lets you call it at decision points inside your own planning loops.

Verdict

The concept is solid and the execution is thoughtful, but the project has only 16 stars and a credibility score of 0.9%, so treat it as an early-stage experiment rather than production-ready infrastructure. The eval suite and documentation are more mature than the adoption metrics suggest, which makes it worth watching--but don't bet a critical architecture decision on it without running your own experiments first.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.