Core Infrastructure

The constitutional layer that keeps the whole monorepo governed, versioned, and deployable.

STATUS: Hardened47 ARCHITECTURE DECISIONSLAST UPDATED: 2026-10-10

Core Infrastructure is the repo-level control plane for the ecosystem. It defines the board workflow, calibration protocol, task numbering, and ADR discipline that every other planet inherits, and runs them over a three-role message bus whose approvals a pre-commit hook checks before any commit lands. Think of it as the operating system for the codebase.

Overview

The Core Infrastructure serves as the foundational monorepo and control plane for the entire ecosystem. It operates as the "Solar System," centralizing all applications while enforcing strict domain isolation, governance rules, and a shared design DNA.

Architecture

Work is coordinated through a three-role message bus whose approvals a pre-commit hook checks before a content commit lands. Traffic is orchestrated by an Nginx API Gateway routing to independent, containerized "Planets" behind a Tailscale zero-trust mesh. The repository utilizes npm Workspaces and Turborepo for high-speed build orchestration, keeping shared logic (like the @mheepeem/ui design system) strictly separated from execution environments.

Problem It Solves

Managing multiple independent projects usually leads to toolchain sprawl, duplicated UI logic, and inconsistent deployment patterns. This core layer solves that by providing a single, governed, and highly reproducible environment where all apps share the same quality gates and infrastructure.

Decisions Worth Knowing

When should an AI agent stop and ask a human? The rules draw the line in two tiers. Decisions about architecture, scope, security, production deploys, or anything beyond an agent's mandate always go to a human, while small choices inside an approved scope, such as a formatting choice or a test fix with an obvious cure, the advising agent settles itself. Messages that need a human are routed by an explicit marker or message type; a second, semantic check was designed and is not built. The standing rule is that over-asking is safe and under-asking is dangerous.

A cost report that was wrong without looking wrong. The report that prices the AI seats' usage had been built from remembered per-token rates instead of the provider's published prices, and a newly adopted model priced at zero without any warning. The fix went to the cause: rates were corrected from the published page, adopting a model now requires checking its live rate, and the report stops with an error on a model it cannot price. The lesson is that a plausible number still needs a source.

One audit, one lesson, several services. A review of the code that handles user-supplied input found the same kind of gap in more than one service. The fixes reused checks the codebase already had instead of adding a shared library, one unused endpoint was deleted rather than defended, and the two safe patterns were written into the coding rules so new code follows them.

Current State

Fully operational and hardened. The Docker compose orchestration and Turbopack builds are active, and the universal ecosystem constitution (Rule 15 Epistemic Standards, Sovereign Workflow) strictly governs all AI and human agent interactions.

MILESTONES

14 milestones
2026