What problem does it solve?
Context engineering prevents agent hallucinations, loss of focus, and convention drift by curating what the agent sees, when it sees it, and how it's structured so outputs remain accurate and aligned with project rules.
Core Features & Use Cases
- Persistent Rules Files: Create and maintain project-wide rules (CLAUDE.md, .cursor rules, GitHub instructions) that agents always load to follow conventions and boundaries.
- Selective Context Loading: Load targeted spec sections, relevant source and test files, and error output per task to keep context focused and under attention budgets.
- Confusion & Session Management: Surface ambiguities, summarize long conversations, and restart sessions when switching major features to avoid stale or conflicting context.
- Use Cases: Starting a new AI-assisted development session, recovering from degraded agent output, switching between subsystems, and onboarding a new project to AI workflows.
Quick Start
Load the project's rules file, the relevant spec section, and the specific source and test files for the task, then ask the agent to propose a prioritized plan that follows project conventions.
