dsh-client-ui-thinking-stats

When working with reasoning-capable models, the volume of thinking tokens and their share of the overall budget often remain opaque numbers buried in a usage log. dsh-client-ui-thinking-stats brings that figure directly into the interface: a cumulative readout in the bottom Composer Dock for the entire session, and a per-turn readout appended to each assistant reply's timing strip, so you can see at a glance how much of every response was spent on internal reasoning.

Each readout displays, left to right, a brain glyph, the thinking-token count, its percentage of all tokens, and its percentage of output tokens, all to one decimal place. Counting prefers the provider-reported reasoningTokens field when available (DeepSeek, OpenAI o-series, Anthropic, and others); if the API returns reasoning text without an explicit count, it falls back to the harness' fixed four-characters-per-token heuristic, so every finalized message contributes a consistent figure regardless of provider.

The plugin is entirely client-side and read-only: it consumes the existing conversation snapshot, adds no service, projection, tool, or RPC call, and depends on no other plugin. When no thinking has occurred, both readouts render nothing, incurring zero DOM or layout cost. Colors follow the --dsw-alias-* theme tokens for automatic light and dark mode support. It is well suited for DSH users who want a lightweight, zero-overhead view of thinking-token share without committing any server-side resources.