30 lines (21 loc) · 1.18 KB

Harness

Local-only AI coding agent (Rust + Gemma 4 via candle / llama-cpp / llama-server). Air-gapped target.

Stack

Main: omo (oh-my-openagent) + ouroboros
Side: —
Final pass: patina --lang en on README / CLI help / setup interview text

Why

zipcode's own thesis is "ride every model" (Gemma + multiple backends) — omo's multi-model philosophy is the same idea, clean dogfooding
Rust + air-gapped verification benefits from ouroboros 3-stage gate (Mechanical → Semantic → Multi-Model Consensus)
Multi-backend compatibility testing benefits from omo's native model diversity

Lane

Lane	Owner
Rust implementation, candle / llama-cpp integration	omo + ouroboros
Backend matrix / compatibility tests	omo (native multi-model)
README / CLI help / setup interview	omo, then patina
Cross-runtime regression on Gemma family	omo

Do NOT

Do not switch primary harness to omx/omo — Korean tone regresses
Do not skip ouroboros seed for pattern changes — anchor verification depends on the spec being explicit
Do not regress the single-static-binary contract — always verify air-gapped install path