The Scurry Lab · Franklin Framework
Harness Engineering
Framework Explorer
Framework Explorer
The outer layer of AI deployment — four distinct architectural layers, each with its own design vocabulary and failure modes.
Dimensions
Global Harness · Overview
The outermost harness layer. Exists before any agent runs, persists after they complete, and applies to all agents at all times. The highest-leverage and highest-risk layer — a change here affects every agent in the system simultaneously.
Dimension 01
Content Architecture
What belongs in the global harness, how is it organized, and what is explicitly excluded?
The global harness holds content that must be consistent across all agents: constitutional bounds, identity and role definitions, lab goals and values, shared knowledge, and operational state. Not everything belongs here — content specific to one agent belongs in the agent harness; content specific to one task belongs in the task harness.
Design questions
- What content must every agent have access to?
- What content is held globally out of convenience rather than necessity?
- What would break if an agent operated without this content?
Failure modes
- Over-populated layer → context bloat; efficiency and coherence degrade
- Under-populated layer → agents lack shared grounding; behavioral inconsistency across agents
- No governance → global layer accumulates stale, contradictory, or orphaned content