System prompt
A system prompt is the instruction text an application places at the start of a language model's context to define its role, rules, and available tools before any user input arrives. In agent systems it carries the operating constraints the agent works under; repository files like AGENTS.md exist precisely to supply project-specific instructions into this layer.
System-level instructions typically take precedence over user turns, which is what makes the layer valuable: it is where an agent's host encodes behavior that should survive the whole session — coding conventions, tool definitions, safety rules, the shape of the task. Agent hosts assemble it from several sources at once: the application's own instructions, user configuration, and files discovered in the working environment.
API producers never control an agent's system prompt directly, and that is the useful insight: what they control is the material agents pull into context alongside it. An AGENTS.md in the SDK repository, an llms.txt that routes to the right pages, and docs concise enough to be included at all are the channels by which a producer's guidance reaches the layer where agent behavior is actually set.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free