dsh-verify-judge: Making Agents Prove Themselves Before Completing Goals

`dsh-verify-judge` is a verification plugin for DeepSeek Harness, designed to address the issue where agents merely self-report “done” without actually passing the tests. It intercepts `update_goal(complete)` calls and enforces running the project’s test suite before closing a goal (supporting automatic detection of `package.json`, `go.mod`, `Cargo.toml`, etc., or a custom command). The plugin also supports turn-boundary checks and failure retries, ensuring the agent keeps working until the task is genuinely completed.