QuoxCORE 2 — core concepts
These five concepts form the vocabulary of QuoxCORE 2. Understanding them is enough to orient any workflow, debug any run, and explain the system to a colleague or an auditor.
Reusable, governed skills that form from practice.
What is a Crystal, how it differs from a Skill or a Prompt, how it forms during a task run, and how tier gating controls promotion.
Specialists with defined authority.
Single agent vs agentic loop, persona-pinned vs tool-using vs browser mode, the mode discriminator, and team composition via team:invoke.
Automation on a canvas. Every step traced.
The node palette, four-way node taxonomy, live execution overlay, versioning, and the workflow library.
Four install paths. One governance model.
OpenAPI, MCP, Repo, and Script install paths; the agent-tool matrix; install-time budget; and plugin durability.
Evidence chains, policy gates, audit export.
AOCL evidence chain, four-scope cost budgets, approval policies, ET-ENFORCE multi-tenant guarantees, and audit trail export.
The governed agentic loop.
A loop-mode Objective handed to a model on a governed fleet dev server: deny-by-default tools, HITL and live gates, per-turn WARD witnessing, self-extension, and an honest maturity map.
Enabled, licensed, active. One server-side gate.
The three states a plugin passes through, how an entitlement attaches to your org, the free-claim device flow, the planned server bands, and the honest ceiling on self-hosted enforcement.