Think Better – 155 decision-science rules for your AI assistant

Hey HN, author here. I noticed a pattern: every time I asked my AI assistant a complex question like "should we migrate to microservices or optimize the monolith?", I'd get the same generic pros-and-cons list. The model has the reasoning capability — it just lacks structured mental models. So I built Think Better: a set of open-source "AI Skills" (following the Anthropic skill format) that inject professional decision science frameworks directly into your local AI setup. How it works under the hood: - The skills contain 16 CSV files with 155 structured knowledge records (decision frameworks, cognitive biases, decomposition methods, mental models) - A Python BM25 search engine ranks the most relevant frameworks for your specific problem type - When you describe a decision or problem, the skill automatically classifies it, recommends the best framework, warns about cognitive biases, and generates a complete evaluation plan For example, if you say "I'm defending our old tech stack because I wrote it", the system triggers Status Quo Bias detection and forces you to calculate the cost of inaction. If you're analyzing why revenue dropped 20%, it builds a MECE Profitability Tree (Revenue = Price × Volume, Costs = Fixed + Variable) and decomposes layer by layer. What it includes: - 10 decision frameworks (Hypothesis Trees, Pre-mortem, Weighted Matrix...) - 12 cognitive biases with auto-detection rules and debiasing remedies - 10 decomposition methods (MECE, Issue Tree, Profitability Tree...) - 12 mental models (First Principles, Inversion, Bayesian Updating...) Everything runs 100% locally. No API keys, no telemetry. The Go CLI is just an installer that mounts the Python/CSV skills into your AI workspace: ./bin/think-better init --ai claude → .claude/skills/ ./bin/think-better init --ai copilot → .github/prompts/ I'd love feedback on: 1. Are there frameworks or biases missing that you'd want? 2. Would you prefer a different skill format (e.g., for Cursor, Windsurf)? 3. Any thoughts on the BM25 approach vs. embedding-based search? GitHub: https://github.com/HoangTheQuyen/think-better