Deep product skill

The presenter demonstrates the "Deep Product" skill, a tool designed to aid in product development ideation and research. The specific problem being addressed is how to best alert users of "Loadout" (a tool for installing and syncing skills and configurations) about updates to installed skills without introducing security risks like prompt injection. The goal is to balance user convenience with safety.

The Deep Product skill analyzes the product's surface area and features, including user types (individual developers, power users), the existing registry system, activity logs, and notifications. It then generates a comprehensive research prompt to explore UX patterns, customer expectations, and safety mechanisms related to updating skills.

The generated prompt is detailed and includes:

• Context about the product, users, and current feature set.
• A description of the specific problem (updating skills).
• Desired state and questions for the research agent, such as establishing UX patterns, update review flows, notification batching, and trust tiers.
• Requests for real-world examples (like password managers), design recommendations, edge cases, anti-patterns, and resource prioritization.

The presenter intends to use this prompt with research tools like Claude and Gemini to gather information and generate a report. A Deep Stack report is briefly shown as an example, illustrating the detailed output including diagrams and code examples.

Deep Product, along with Deep Stack and Interview Me, are presented as tools to facilitate research and ideation, helping users to avoid building the wrong thing. Deep Research is particularly useful for exploring existing solutions and patterns in the real world, even within complex open-source projects.