ai-workspace-reference

An opinionated operating system for agentic development and agentic automation, built on plain files and folders.

This repository is a public reference implementation of the workspace architecture described in team-lib/context/indexed/workspace-reference.md. It demonstrates the patterns, conventions, and directory structure without any proprietary content.

What This Is

The Pvragon AI Workspace solves four problems that make AI agents unreliable over long-running work:

• Context amnesia — agents forget everything between sessions
• Stochastic behavior — agents do the same task differently each time
• No identity — agents have no persistent context about your business
• Compounding errors — without guardrails, small mistakes snowball

The solution is a four-layer hierarchy with clear boundaries, registry-driven execution, and a two-axis tiered memory system with ranked retrieval.

Architecture at a Glance

Dependencies flow downward only. Projects consume from libraries. Libraries never depend on projects.

Core Patterns

DOE (Directive–Orchestration–Execution)

Every task separates what (directives) from who decides (the AI reasoning loop) from how (deterministic scripts). Reliability comes from pushing repeatable steps into tested executions.

Memory: two orthogonal axes

Memory is layered on two independent axes that cross rather than stack — every memory has a coordinate on both.