context-engineering

You manage AI agent context by deliberately curating what the agent sees and when it sees it through rules files, hierarchical spec loading, and prioritized source files, preventing hallucinations from insufficient context and drift from excessive context.

Context engineering is the practice of curating what an AI agent sees, when it sees it, and how it is structured. It uses rules files and hierarchical loading of specs and source files to govern agent context and prevent output quality decline.

Create a CLAUDE.md rules file at your project root and load it before starting a new session to persist project-wide guidelines, govern agent access boundaries, and apply rules consistently across coding sessions.

You should adjust agent context when starting a new coding session, when agent output quality declines, when switching between tasks, and when configuring project-wide rules and per-task source file loading.

Load relevant spec sections per feature or task to guide planning and implementation without loading the entire document, keeping the agent aligned with current work and maintaining focused output.

AI agents lose focus during long sessions due to unmanaged context and session history drift. Managing conversation history and context prevents drift and maintains focus across switching tasks.