AI Native Engineering OS

AI-Native Engineering OS

ClawAI treats engineering context as a first-class system. Human contributors and coding agents are routed through the same canonical knowledge layer.

Authority stack

1. Root policy/router files such as CLAUDE.md.
2. Hard blockers in rules/00-non-negotiable-rules.md.
3. Canonical structure in context/architecture-map.md.
4. Task-specific rules, skills, context, memory, service-local AGENTS.md, and generated manifests.
5. Implementation code and tests.

The generated .ai layer

The repository currently contains 33 files under .ai/. The manifests include:

• API endpoints
• data ownership
• Docker services
• environment variables
• event graph
• frontend routes
• governance
• hashes/freshness
• i18n
• Nginx routes
• packages
• permissions
• ports
• Prisma models
• RabbitMQ events
• repository metadata
• services
• tests
• workspace dependency graph
• workspaces

The pack layer includes task-oriented bundles for provider connectors, authentication/security, billing/payments, chat streaming, database migrations, documentation, frontend features, infrastructure, model routing, RabbitMQ events, and workspace connectors.

Context-first workflow

Before engineering work, the repository expects task classification/context resolution so the contributor knows:

• affected workspaces;
• rules that govern the change;
• skills/runbooks to execute;
• likely files and tests;
• generated artifacts that must be refreshed;
• relevant pitfalls and prior lessons.

See Context-Layer, Skills-Catalog, Rules-Catalog, Engineering-Memory, and Stack-and-Toolchain.