context.md vs SKILL.md: Stop Pasting Product Facts Into Every Prompt

The separation of concerns for agent operations is crucial for maintaining consistent, accurate, and efficient interactions. Instead of pasting product facts into every prompt, it is recommended to have two distinct files: one for the procedure (SKILL.md) and another for the facts (context.md).

SKILL.md should contain the procedure, when to use it, the goal, inputs to collect, process steps, output shape, and anti-patterns. This file should remain stable and rarely need changes. It should not include any product-specific details like company name, ICP paragraphs, exact prices, or current offers.

On the other hand, context.md should hold the mutable facts that change frequently, such as product name, audience, tone, must-not rules, current offer, and canonical links. This file should be updated weekly or whenever there are changes in the product.

By keeping these two files separate, agents can focus on following the established procedure without being distracted by constantly changing facts. The skill should read the context file, not contain it. This way, the skill remains thin and durable, while the context adapts to the evolving product and market conditions.

The recommended approach is to have a minimal context.md template under ~40 lines, detailing the essential information that changes frequently. The skill should be triggered with the skill name and the day-of payload, without any additional ICP essays or pricing details. This ensures a consistent and reliable interaction process while allowing the context to adapt to the changing product landscape.