← back to projects
TOME

AI Harness Orchestration Platform — the methodological backbone of everything ZPAK ships. The harness is the product.

OPERATIONAL HARNESS v2.0 79 FILES

Tome is ZPAK's AI orchestration engine — built on the principle that the harness is the product. It manages the coordination of agents, skills, context layers, and system prompts through a structured, version-controlled harness architecture.

Where most AI deployments are ad hoc — individual prompts, one-off sessions, disconnected tools — Tome imposes rigorous structure on the entire intelligence stack. Every agent has a defined role. Every skill is version-controlled. Every context layer is intentional and inspectable.

Tome operationalizes the RMCC (Recursive Metacognitive Cognitive Cycle) protocol — a five-pass quality gate that transforms raw AI output into defensible, high-density artifacts. Every output produced by the ZPAK stack runs through RMCC before it's considered complete.

The harness sits above any single model or provider. It routes cognition to the right layer:

  • Heavy reasoning and analysis routed to Claude Pro
  • Cloud inference fallback via DeepInfra for cost-sensitive tasks
  • Local tasks handled by Ollama on the research-zen-WST workstation
  • Agent framework layer powered by Mastra.ai

The Tome harness bundle — 79 files across five tiers — was produced by a multi-threaded generation operation processing 421 corpus items. It serves as the living operating manual for the entire ZPAK intelligence stack.

SOUL tier
Operator identity, values, and cognitive profile
SEED tier
Base knowledge and domain context
CTX tier
Active context layers (12 domains)
AGENT tier
Agent definitions and role specifications
SKILL tier
Versioned skill modules and prompt libraries
Total files
79 harness files · 421 corpus items processed
  • DONEHarness v2.0 bundle delivered and operational
  • DONERMCC five-pass quality gate implemented
  • DONEMastra.ai agent framework integration
  • DONEMulti-provider routing (Claude / DeepInfra / Ollama)
  • WIPHarness v3.0 planning — expanded SKILL tier
  • NEXTAutomated harness regeneration pipeline
  • NEXTSENTINEL integration for briefing synthesis