Idea Refinery

Problem It Addresses

This skill targets several common failure modes in LLM-generated creative work: multiple options look distinct but remain local variations of the same starting point; a plan “seems reasonable” without checking budget, cost, schedule, or infrastructure; and AI recommendations can package fragile judgments as confident conclusions.

How It Works

Idea Refinery turns creative refinement into an auditable workflow:

• Beginner mode: avoids role names, uses 1-2 clarification rounds, three independent paths, quick risk screening, and one executable first step for fast direction-finding.
• Standard mode: the default mode, organized around Dreamer, Builder, and Opponent roles to complete clarification, divergence-axis extraction, three-dimensional critique, an Operator action tuple, code feasibility checks, a human decision node, and reality-constraint review.
• Deep mode: activates all roles, an expert, high-intensity confrontation, debate, and a value-validation panel for strategic or complex projects.

The core mechanism is divergence-axis driven: identify the most consequential uncertain variable, take three extreme values as initial conditions, and derive path A/B/C separately to reduce natural convergence. Then the Opponent flags market, cost, and feasibility risks, while the Operator outputs an executable plan in the form who + what + first step.

The skill also requires AI to generate pure standard-library Python code for rough estimates of revenue, cost, resource gaps, or scheduling conflicts, using execution results to mark feasibility. Finally, a human decision node presents path assumptions, resource needs, estimated outcomes, and risks side by side, explicitly avoiding AI recommendation.

Scope and Caveats

It fits product planning, marketing campaigns, management decisions, technology selection, and career planning where creative thinking and critical thinking are needed. It is not intended for pure execution tasks, factual Q&A, real-time data retrieval, or emotional support. The code simulation is only a range estimate, not a precise forecast; legal and compliance conclusions should still be verified against the actual context.