Subagent model control for DeepSeek Harness. A Jev classifier labels each turn (task class, effort, blast radius, risky flags), and the subagent's own request is rewritten to the provider, model and reasoning effort that label maps to, enforced on the agent/request waterfall. Also: a per-session inherit/auto/manual selector with a status badge, one prompt per role injected as a message ahead of the task, a right-sidebar tab listing each role's model and recent dispatches, credential-shaped text redacted before it leaves the machine, and an offline replay tool for past verdicts.
This plugin has not published an npm package, so the store does not offer an install command yet. Use the repository link below to view the source.