cARL (Cognitive Agent Runtime Layer) is a repository-native control layer for AI-assisted development. It defines persistent instructions, boundaries, and operating constraints so coding agents behave consistently over time.
What it does
- Provides committed policy and behavior files for agent execution
- Preserves engineering intent as versioned project context
- Reduces drift in how autonomous coding tasks are handled
Why it matters
Agent outputs are only as reliable as the rules around them. cARL makes those rules explicit, reviewable, and durable across sessions and contributors.