Jevvy
A plugin collection for coding agents, powered by TypeSafe's System One model, that automatically approves harmless shell permission requests in OpenCode and Claude Code while leaving any uncertain request to the agent's normal approval flow.
Jevvy Permissions reviews shell commands that a coding agent is about to run and asks for permission for, evaluating each pending request as a complete command against fast probabilistic judgments from TypeSafe's System One model. It only overrides a request that its own agent has already left unresolved: existing allow and deny rules stay final, and any command visibly configured to always ask a human is left as a human prompt.
It auto-approves the current action only when every check on it passes, and otherwise abstains, leaving the agent's own permission flow unchanged rather than blocking or forcing a decision. Configuration lives in one global file that project repositories cannot override, and it can use TypeSafe AI, OpenCode Zen, OpenRouter, the Vercel AI Gateway, a project called Laya, or a fully custom HTTP endpoint as its judgment provider.