AI Agent Hub
Back to plugins
🖥️

dsh-turn-fold

Client Updated 2026.09.14

Run the following command in DeepSeek Harness:

dsh plugin install ruisenbai/dsh-turn-fold

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install ruisenbai/dsh-turn-fold to add this local Web profile layer in the DeepSeek Harness CLI (repo: https://github.com/ruisenbai/dsh-turn-fold), then refresh the running Web page to activate the bundle.

About this plugin

In the DeepSeek Harness Web UI, a single conversation can interleave many reasoning steps and a dozen tool calls, stretching the page so far that the final reply gets buried. dsh-turn-fold groups the reasoning and tool calls within each turn into a single collapsible step bar: while the turn is active it shows the current tool or a short reasoning preview, and once the turn completes it folds automatically, leaving only the final reply in view. Clicking the bar reveals the original reasoning and tool cards layer by layer; the summary strip also surfaces tool icons, file counts, read/edit/search tallies, and added/removed line totals from persisted diff metadata.\n\nThe plugin ships a bilingual (English/Chinese) settings card where each metric—duration, first-token latency, token usage, generation speed, cache hit rate, step count—has its own toggle, and numeric checkpoints can be saved locally so they survive browser or Host restarts. Animations respect the system reduced-motion preference, focus is preserved during expand/collapse, and the mobile layout is accounted for.\n\nIt introduces no model input, no extra tools, and no KV-cache side effects; it is purely a browser-side presentation layer. It is well suited to DeepSeek Harness users who run multi-step reasoning, file editing, and search sessions in the Web UI and prefer a compact view that surfaces the conclusion first, with process details available on demand.

Use Cases

  • Long multi-step Web conversations where process cards bury the final reply
  • Reviewing per-file edit diffs and tool call counts after a turn completes
  • Preferring a compact, conclusion-first UI with on-demand reasoning detail

Best For

  • DeepSeek Harness Web users running multi-turn conversations daily
  • Developers who heavily use file editing and search tool calls
  • Power users who track generation speed, token usage, and cache hit rate