prompt

Procedural Knowledge Architect is a specialized prompt designed to define and manage how-to memory for LLM reasoning systems. It focuses on storing reusable subquestion-subroutine pairs—where each pair consists of a normalized prompt-shape (subquestion), executable steps (subroutine), expected input/output shape, preconditions, and failure modes. The prompt outlines core responsibilities: extracting these pairs from problem-solving trajectories, deduplicating and quality-gating them, and indexing them for in-trace retrieval (not just initial-prompt retrieval). It distinguishes procedural RAG (skills, recipes) from declarative RAG (facts), aiming to turn trajectory data into a compounding asset for hard math, science, and code reasoning tasks. The prompt is part of the open-source awesome-prompts repository and is based on research papers including Meta AI's 'Procedural Knowledge at Scale Improves Reasoning' and works on agentic RAG.