OpenCode powers Catan clone with local LLM

This is a strong practical demo for terminal-based coding agents: not a toy script, but a rules-heavy, stateful game shipped fast with minimal prompting.

• –The project shows OpenCode can handle multi-file gameplay logic, UI flow, and multiplayer coordination when the spec is well defined
• –Local-model coding is past the “can it write code?” stage and into “can it hold a complex product shape together?” territory
• –The remaining bug about visible hands is the kind of domain-specific edge case that agents often miss unless humans force privacy requirements explicitly
• –The real story is less the mystery model and more the workflow: OpenCode + a good spec can compress a two-person prototype into a weekend build
• –If this claim holds up in a live demo, it is a useful signal for teams evaluating local-first coding stacks