dsh-tool-highlight

In DeepSeek Harness, when the model reads a Python file or executes a shell script, the terminal output is typically a wall of plain, uncolored text: keywords, strings, and comments all blend together, making extended reading tedious. dsh-tool-highlight tackles exactly this problem by adding layered syntax highlighting to read code output and bash / pwsh command output, assigning distinct colors to keywords, strings, numbers, function names, and comments in the VS Code Dark+ palette, with automatic light/dark theme switching that follows DSH global tokens.

The plugin follows a "color only what you recognize" philosophy: read cards detect language by file extension (.py, .js, .ts, .json, .ps1, .sh, etc.), while command output is classified by content heuristics: JSON-structured text gets JSON coloring, code-like content gets syntax coloring, and tables or logs are left untouched to avoid noisy over-highlighting. Each card also shows an exit-code status bar (green for success, red for non-zero exit) and supports a default-collapsed single-line title that expands on click; running and error states auto-expand so you never miss live output.

The entire plugin is pure front-end rendering: it adds no new conversation content, modifies nothing, calls no model, and incurs zero token cost. The tokenizer is a self-written lightweight regex engine with zero runtime dependencies. It exclusively handles the read, bash, and pwsh tool cards (priority -2), coexisting cleanly with dsh-better-tool-ui, which manages the remaining tools. Ideal for developers who frequently inspect code reads and command output in the DSH web interface.