writing-feature-spec

To write an agent-ready feature specification, you structure product requirements into context layers (L1-L4) for efficient information loading. This progressive disclosure method ensures AI agents can parse functional requirements and user stories without being overwhelmed by unstructured text.

The best way to format user stories for spec-driven development is using a structured Markdown template with Given-When-Then acceptance criteria. This clearly defines functional requirements and non-goals, ensuring unambiguous execution by AI agents.

Progressive disclosure in product requirements documents works by organizing information into distinct layers (L1-L4). This context-layered approach allows AI coding agents to efficiently load only the necessary specification details, reducing token usage and improving implementation accuracy.

Standard PRD templates often lack the structured context layers required for AI agent spec-driven development. You need an agent-optimized feature specification format that separates functional requirements, user stories, and non-goals into distinct layers for efficient machine parsing.

A feature spec to start spec-driven development should include functional requirements, user stories in Given-When-Then format, acceptance criteria, and non-goals. Organizing these into a structured Markdown file like `specs/prd-{feature-name}.md` initiates clear AI agent execution.