i-have-adhd

i-have-adhd is a coding-assistant skill that reshapes an AI agent's output for a reader with ADHD. One SKILL.md file encodes ten output rules grounded in five facts about ADHD cognition: small working memory, the knowing-doing gap, cold-start paralysis, time blindness, and dopamine scarcity. The rules: lead with the next action, number multi-step work, end with one concrete next step, suppress tangents, restate state every turn, give specific time estimates, make wins visible, report errors matter-of-factly, cap lists at five, and forbid all preamble, recap, and closers. The rules persist for the whole session until the reader says "stop adhd mode".

The repo is more than a prompt. The same ruleset is packaged for ten-plus harnesses — Claude Code, Codex, Pi, Gemini CLI, Kimi, Qwen, Cursor, GitHub Copilot, Zed, Hermes — via small per-harness manifests, with three always-on routes layered on top. The deepest integration is a Pi extension (extensions/i-have-adhd.ts) that injects the ruleset once per session as a typed message, re-injects it after compaction, and restores on/off state from saved session state. A Claude Code SessionStart hook makes the mode always-on behind an opt-in flag file (~/.claude/.i-have-adhd-always).

The project also ships a paired eval harness (scripts/run_evals.py, evals/) that treats output shaping as a measurable quality regression: baseline vs. candidate responses judged blind on a weighted rubric (correctness 35%, autonomy 25%, actionability 20%, safety 10%, concision 10%), gated behind a release check that the candidate must beat baseline on.