DeepSeek Harness Plugin: Using the Jev Decision Model to Streamline Output, Screen for Injections, and Recommend Skills

`dsh-jev-tools` is a security plugin for DeepSeek Harness that uses the Jev model to filter and evaluate content before it enters the context. Its core capabilities include simplifying tool outputs (such as `read` and `grep`), screening malicious instructions, recommending skills, and actively invoking `jev_ask` and `jev_gate`. The plugin defaults to fail-open, so failures do not affect tasks except for `jev_gate`. Installation requires specifying a profile, and after configuring the TypeSafe API Key, it takes effect automatically, making it suitable for scenarios that require fine-grained control over context and security.