adhd

ADHD is a skill designed for coding agents, built on the Claude and Codex Agent SDK, to address premature convergence in autoregressive reasoning. Unlike linear Chain-of-Thought or traditional Tree-of-Thought, ADHD spawns N isolated reasoning processes under deliberately distorted cognitive frames with zero shared context during divergence, then runs a separate critic pass to score, cluster, prune traps, and deepen survivors. This two-phase loop (diverge and focus) is mechanical, using separate LLM calls with opposite system prompts. Installation is simple: `npx skills add UditAkhourii/adhd` auto-detects agents like Claude Code, Cursor, Codex, and more. It can be invoked explicitly with `/adhd "problem"` or auto-triggered on ideation intents. A CLI (`adhd-agent`) and a TypeScript library are also available. ADHD is particularly useful for design decisions, fuzzy debugging, naming, API surface design, strategy, and prompts like "give me a few ways to…". It has been adopted by 17+ projects, including repowire, mstack, and han. Independent evaluations show significant gains in breadth, novelty, and trap detection compared to single-shot baselines, though at a cost of ~2.3× time and ~1.9× output. The project includes comprehensive documentation, a preprint, and an active community on Discord.