moai-foundation-thinking

Use the diverge-converge framework: gather requirements, generate 20-50 ideas without filtering, cluster them into 4-8 themes, score clusters with weighted criteria like feasibility and impact, then document the top 3-5 candidates with validation plans.

Use the First Principles five-phase process: audit assumptions, decompose the problem to fundamental truths, generate 5-10 alternatives, score trade-offs across correctness, performance, and maintainability, then check for cognitive biases like anchoring and sunk cost.

Apply the 7-step critical evaluation: restate the claim, assess evidence quality, detect logical fallacies, expose unstated assumptions, generate alternative explanations, check internal contradictions, and verify the burden of proof matches the claim's magnitude.

Use deep questioning for features, architecture decisions, and unfamiliar domains where requirements are unclear. Skip it for simple tasks touching 1-2 files. The six layers progress from surface understanding through root cause, constraints, users, solution boundaries, and validation risks.

Yes, the ultrathink keyword triggers extended reasoning on Opus 4.7 and later models, complementing the creative frameworks. Adaptive Thinking allocates reasoning depth automatically based on task complexity, and ultrathink forces deeper reasoning when needed.

These frameworks add process overhead unsuited for trivial tasks like quick bug fixes or configuration changes. They also depend on honest input: shallow answers during deep questioning or fewer than three alternatives during divergence undermine the output quality.