ACE — AI Context Environment

As Circuitte's monorepo grew across multiple apps and packages, AI assistants drifted — re-reading the wrong files, missing conventions and producing inconsistent changes. Project knowledge was scattered across READMEs, chat history and people's heads, and whatever was written down went stale within weeks.

The cost was concrete: every AI-assisted task started with re-discovery, architectural decisions were re-litigated instead of referenced, and the quality of generated changes depended on which files happened to be in context. Scaling the team — human or AI — meant scaling the confusion.

ACE (AI Context Environment) structures project knowledge into a four-layer pyramid: root rules, package rules, category summaries and detailed design docs. An AI assistant descends only as deep as the task requires, so context stays small, current and relevant.

The pyramid is kept honest by machinery, not discipline alone: a live manifest with versioning, an in-app dashboard, freshness thresholds with smoke tests, and a technical-decision-record (TKD) workflow that captures architectural choices where the next session will find them. Specialized sub-agents — audit, context-refresh, inventory, security — maintain the environment on schedule.

The result is an engineering organization where AI sessions start informed: conventions are followed on the first attempt, decisions persist across sessions, and the context itself is tested like code.