dsh-thinking-counter

When using DeepSeek Harness for complex reasoning, you might wonder how often the AI leans on phrases like "let me" or "we should." The dsh-thinking-counter plugin turns this curiosity into observable data. It listens to the thinking process of all conversations in real-time, automatically counting the frequency of specific phrases and making the AI's "linguistic habits" transparent.

Its core strength lies in its comprehensive and detailed approach. Beyond counting standard reasoning blocks, it allows you to freely add and group detection words, with individual case-sensitivity settings for each. The most distinctive feature, the "Whale Pupil Detection" function, transforms dry statistics into a fun "intelligence" game: it presents results through adorable whale-girl illustrations and humorous judgments (like "Liang Sheng Possessed" or "Slacking Off") based on the ratio of positive words to total hits.

This plugin is especially suited for power users of DeepSeek Harness, AI behavior researchers, and enthusiasts intrigued by model reasoning processes. Developers can use it to debug and observe the model's word-choice tendencies during long-chain thinking, while researchers gain intuitive linguistic usage statistics. Of course, if you simply want to add some entertainment to your workflow, using the "Fat Fish" judgment as a mood reliever, it serves perfectly for that too.