opencode-harness

Set up and use the OpenCode 'zen' model harness in any project — route agent work to cheaper models by capability (fast=kimi-k2.6, reasoning-heavy/code-generation=glm-5.2) to save Claude's token budget, with Claude orchestrating and validating. One-command setup writes ~/.config/vinay-agent/config.toml and tests both models end-to-end; ask.sh calls a profile; /opencode-harness command included. Built by dogfooding — measured routing findings included. Use when integrating OpenCode, offloading reasoning/codegen to a cheap endpoint, or deciding which model a task should go to.