UditAkhourii/adhd

What it solves

ADHD addresses "premature convergence" in autoregressive reasoning, where AI agents tend to anchor on their first idea and fail to explore a wide range of creative or non-obvious solutions. It prevents the narrow, "textbook" answers typical of single-shot or linear Chain-of-Thought reasoning by forcing the model to explore multiple isolated perspectives before converging on a final answer.

How it works

The system uses a two-phase mechanical loop with a strict separation between generation and evaluation:

1. Diverge: The project spawns N parallel, isolated reasoning processes. Each process is given the problem and a specific "cognitive frame" (a vantage prompt) but is forbidden from evaluating the result. Because these branches share no context, they cannot anchor on each other's ideas.
2. Focus: A separate critic agent scores every generated idea based on novelty, viability, and fit. It then flags "traps" (ideas that seem good but are broken), clusters ideas by angle, and deepens the top-K survivors into detailed sketches.

Who it’s for

It is designed for developers and AI agent users who need high-quality ideation for complex tasks such as design decisions, API surface design, fuzzy debugging, naming, and strategic planning.

Highlights

• Isolated Reasoning: Prevents anchoring by ensuring divergent branches have zero shared context.
• Trap Detection: Specifically identifies and explains why certain seductive but flawed ideas should be avoided.
• Agent Agnostic: Can be installed as a skill for various agents including Claude Code, Cursor, and Windsurf.
• Proven Novelty: Demonstrates significantly higher novelty and breadth scores compared to single-shot baselines in engineering benchmarks.