Your AI gives everyone the same answer — once a model picks a direction it stops exploring, so the one thing a paid plan can't sell you is an idea nobody else got. The free, MIT-licensed ADHD skill (github.com/UditAkhourii/adhd, 4,000+ stars) attacks exactly that: /adhd fans your question out to five fully isolated agents, each locked into a different cognitive frame, then a critic scores every idea, flags the traps with reasons, and deepens the three strongest. On Aug 30, 2026 we put it on trial — an independent, pre-registered blind test: two real engineering problems (a SQLite desktop-app migration; a Node exit-0 heisenbug), plain single-shot vs the full ADHD loop, judged blind on the repo's own four axes with positions swapped as a control. It won breadth 9.0 vs 6.5, novelty 9.0 vs 4.5, and trap detection 8.5 vs 5.5 — and LOST actionability 6.0 vs 9.0 in both rounds, at roughly 8× the tokens (the repo's own table claims that axis too; our blind judge reversed it). This card is the whole discipline: PAGE 2 — the use-it-or-skip-it decision card, the skill's own pre-flight gate verbatim, and the honest cost math. PAGE 3 — all 15 cognitive frames with their verbatim vantage prompts. PAGES 4–5 — both case-study receipts (fork-don't-mutate against Time Machine restores; the completion ledger that makes exit code 0 earned) plus the full blind scorecard, the reversal, and the honest limits. PAGE 6 — the trap gallery: the four attractive ideas the skill flagged against itself. Every number is a committed receipt from our own run. Independent guide from Hyperautomation Labs — not affiliated with the ADHD project.

By submitting, you'll be subscribed to the Hyperautomation AI Report newsletter — one practical AI email per week. Unsubscribe anytime.

Free. No spam. Unsubscribe anytime.