adhd

Why we included this project

A common failure mode in agent sessions is committing to the first plausible answer before anything better has a chance to surface. ADHD attacks that anchoring by spawning several isolated reasoning branches at once, each pushed through a deliberately different cognitive frame, such as a regulator, a speedrunner, a 10-year-old, or a $0 budget, with no shared context while they diverge. A separate critic pass then scores the branches, clusters similar ideas, prunes obvious traps, and spends extra effort deepening whatever survives. For teams running Claude Code or Codex on open-ended work like design decisions, API surface design, naming, or fuzzy debugging, this turns a single-threaded brainstorm into a structured exploration you can inspect and steer, and it surfaces non-obvious picks a linear chain would never consider. It installs as a skill you invoke on demand, so it slots into an existing agent workflow rather than forcing a rewrite.