AI Agent Hub
Back to plugins
🖥️

dsh-stats-board

Client Updated 2026.08.18

Run the following command in DeepSeek Harness:

dsh plugin install PastSheep/dsh-stats-board

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

Run "dsh plugin install PastSheep/dsh-stats-board" in your terminal; the full source is available at https://github.com/PastSheep/dsh-stats-board . Restart DeepSeek Harness after installation to see the Stats tab in your conversation view.

About this plugin

DeepSeek Harness Web ships with Conversation and Trace views built for step-by-step replay, but getting a single glance at session-level latency distribution, token flow, tool-call frequency, and cache hit rate still means combing through logs. dsh-stats-board registers a third Stats tab right beside those two and places an empty component on the composer-dock slot to suppress the built-in stats line, pulling the numbers into a proper visual layer.

Open the board and floating KPI cards lay out total turns, LLM wall-time, average TTFT, output speed, cache hit rate, and context pressure. A hand-rolled SVG donut chart switches between a context-projection mode (system / tools / messages) and a billing-bucket mode (uncached input / cache read / cache write / output). Blue-and-purple bar charts show per-turn input and output over the last 16 rounds, while a cumulative output line traces the token trajectory. Two compact tables at the bottom list per-turn details and a tool-call leaderboard sorted by frequency and total latency. Everything is written in plain JavaScript with React.createElement and inline SVG; there is no build step, so the board renders the moment the plugin loads.

If you tune prompts in DeepSeek Harness every day, watch TTFT, track token budgets, or need to walk a teammate through session health checks, this dashboard eliminates the log-reading ritual and puts the metrics that matter directly in front of you.

Use Cases

  • Quickly review session latency, tokens, and cache metrics
  • Toggle token composition between context projection and billing buckets
  • Walk a teammate through a session health overview

Best For

  • Developers tuning prompts daily
  • Engineers tracking TTFT and output speed
  • Ops teams calculating token budgets