dsh-prism

After DeepSeek Harness shipped, the community criticism was not about missing features but about the steep learning curve: layered YAML configs, plugin assembly, and effect components leave non-developers staring at a wall of jargon. dsh-prism tackles that exact tension. It gives you two modes on the same interface—plain-language for newcomers, full native for power users—with a single toggle and zero interference.

The core capability is the two-mode system. In Simplified mode, every tool call that happens before the model's final reply in a single user turn is collapsed into one group. Collapsed, it shows a one-line summary (tool count, reasoning count, status); expanded, it reveals a documented panel where each tool gets a plain-language row with a category icon, an action summary, and a redacted delivery-document detail view. Twenty-one advanced tools are folded into summary lines by default to cut noise. Native mode registers nothing at all — the product renders itself untouched, pixel for pixel. On top of that, the plugin ships data redaction (paths show only filenames, key-shaped strings are replaced with placeholders, sensitive parameter names are never displayed), a full 33-tool plain-language rule table, bilingual UI that follows DSH's language setting in real time, and full light/dark theme support via official design variables.

It is built for two audiences: newcomers who want to use DeepSeek Harness but are intimidated by the terminology and configuration, and experienced users who occasionally want a quick plain-language scan of tool execution summaries. The plugin is a pure presentation-layer add-on; model inputs and outputs are untouched, a page refresh returns you to Native mode, and uninstalling leaves no residue.