context-engineering

Inconsistent AI agent outputs occur when context is not deliberately curated. Applying context engineering delivers the right information at the right time, preventing hallucinations from too little context and loss of focus from too much.

Context engineering is the practice of deliberately curating what an AI agent sees, when it sees it, and how it is structured. It solves inconsistent outputs by applying a defined context hierarchy and loading strategy.

Structure context from most persistent to most transient: Rules Files, Spec or Architecture Docs, Relevant Source Files, Error Output or Test Results, and Conversation History. This hierarchy maintains output quality across tasks.

Configure a rules file like CLAUDE.md to persist project-wide conventions across sessions. Combine this with a defined context hierarchy and per-task specs to guarantee deterministic, high-quality outputs.

Avoid context flooding by loading only relevant source file sections per task. Establish defined trust levels for loaded context to ensure reproducibility and prevent the agent from losing focus due to excessive information.

Use rules files to establish persistent project-wide conventions across all sessions, and use task specs for transient, specific instructions. Combining both within a structured context hierarchy ensures deterministic agent governance.